56
1 Transport Layer 3-1 Chapter 3 Transport Layer Computer Networking: A Top Down Approach Featuring the Internet, 3 rd edition. Jim Kurose, Keith Ross Addison-Wesley, July 2004. Transport Layer 3-2 Chapter 3: Transport Layer Our goals: understand principles behind transport layer services: multiplexing/demultipl exing reliable data transfer flow control congestion control learn about transport layer protocols in the Internet: UDP: connectionless transport TCP: connection-oriented transport TCP congestion control

Chapter 3 Transport Layer - University of Utah

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

1

Transport Layer 3-1

Chapter 3Transport Layer

Computer NetworkingA Top Down ApproachFeaturing the Internet3rd editionJim Kurose Keith RossAddison-Wesley July2004

Transport Layer 3-2

Chapter 3 Transport LayerOur goals understand principles

behind transportlayer services multiplexingdemultipl

exing reliable data transfer flow control congestion control

learn about transportlayer protocols in theInternet UDP connectionless

transport TCP connection-oriented

transport TCP congestion control

2

Transport Layer 3-3

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-4

Transport services and protocols provide logical communication

between app processesrunning on different hosts

transport protocols run in endsystems send side breaks app

messages into segmentspasses to network layer

rcv side reassemblessegments into messagespasses to app layer

more than one transportprotocol available to apps Internet TCP and UDP

applicationtransportnetworkdata linkphysical

applicationtransportnetworkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysicalnetwork

data linkphysical

logical end-end transport

3

Transport Layer 3-5

Transport vs network layer

network layer logicalcommunicationbetween hosts

transport layer logicalcommunicationbetween processes relies on enhances

network layer services

Household analogy12 kids sending letters

to 12 kids processes = kids app messages = letters

in envelopes hosts = houses transport protocol =

Ann and Bill network-layer protocol

= postal service

Transport Layer 3-6

Internet transport-layer protocols

reliable in-orderdelivery (TCP) congestion control flow control connection setup

unreliable unordereddelivery UDP no-frills extension of

ldquobest-effortrdquo IP services not available

delay guarantees bandwidth guarantees

applicationtransportnetworkdata linkphysical

applicationtransportnetworkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysicalnetwork

data linkphysical

logical end-end transport

4

Transport Layer 3-7

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-8

Multiplexingdemultiplexing

application

transport

network

link

physical

P1 application

transport

network

link

physical

application

transport

network

link

physical

P2P3 P4P1

host 1 host 2 host 3

= process= socket

delivering received segmentsto correct socket

Demultiplexing at rcv hostgathering data from multiplesockets enveloping data with header (later used for demultiplexing)

Multiplexing at send host

5

Transport Layer 3-9

How demultiplexing works host receives IP datagrams

each datagram has sourceIP address destination IPaddress

each datagram carries 1transport-layer segment

each segment has sourcedestination port number

host uses IP addresses amp portnumbers to direct segment toappropriate socket

source port dest port

32 bits

applicationdata

(message)

other header fields

TCPUDP segment format

Transport Layer 3-10

Connectionless demultiplexing

Create sockets with portnumbers

DatagramSocket mySocket1 = newDatagramSocket(12534)

DatagramSocket mySocket2 = newDatagramSocket(12535)

UDP socket identified bytwo-tuple

(dest IP address dest port number)

When host receives UDPsegment checks destination port

number in segment directs UDP segment to

socket with that portnumber

IP datagrams withdifferent source IPaddresses andor sourceport numbers directedto same socket

6

Transport Layer 3-11

Connectionless demux (cont)

DatagramSocket serverSocket = new DatagramSocket(6428)

ClientIPB

P2

client IP A

P1P1P3

serverIP C

SP 6428DP 9157

SP 9157DP 6428

SP 6428DP 5775

SP 5775DP 6428

SP provides ldquoreturn addressrdquo

Transport Layer 3-12

Connection-oriented demux

TCP socket identifiedby 4-tuple source IP address source port number dest IP address dest port number

recv host uses all fourvalues to directsegment to appropriatesocket

Server host may supportmany simultaneous TCPsockets each socket identified by

its own 4-tuple Web servers have

different sockets foreach connecting client non-persistent HTTP will

have different socket foreach request

7

Transport Layer 3-13

Connection-oriented demux(cont)

ClientIPB

P1

client IP A

P1P2P4

serverIP C

SP 9157DP 80

SP 9157DP 80

P5 P6 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

Transport Layer 3-14

Connection-oriented demuxThreaded Web Server - only 1process

ClientIPB

P1

client IP A

P1P2

serverIP C

SP 9157DP 80

SP 9157DP 80

P4 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

8

Transport Layer 3-15

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-16

UDP User Datagram Protocol [RFC 768]

ldquono frillsrdquo ldquobare bonesrdquoInternet transportprotocol

ldquobest effortrdquo service UDPsegments may be lost delivered out of order

to app connectionless

no handshaking betweenUDP sender receiver

each UDP segmenthandled independentlyof others

Why is there a UDP no connection

establishment (which canadd delay)

simple no connection stateat sender receiver

small segment header no congestion control UDP

can blast away as fast asdesired

9

Transport Layer 3-17

UDP more often used for streaming

multimedia apps loss tolerant rate sensitive

other UDP uses DNS SNMP

reliable transfer over UDPadd reliability atapplication layer application-specific

error recovery

source port dest port

32 bits

Applicationdata

(message)

UDP segment format

length checksumLength in

bytes of UDPsegmentincluding

header

Transport Layer 3-18

UDP checksum

Sender treat segment contents as

sequence of 16-bitintegers

checksum addition (1rsquoscomplement sum) ofsegment contents

sender puts checksumvalue into UDP checksumfield

Receiver compute checksum of

received segment check if computed checksum

equals checksum field value NO - error detected YES - no error detected

But maybe errorsnonetheless More laterhellip

Goal detect ldquoerrorsrdquo (eg flipped bits) intransmitted segment

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

2

Transport Layer 3-3

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-4

Transport services and protocols provide logical communication

between app processesrunning on different hosts

transport protocols run in endsystems send side breaks app

messages into segmentspasses to network layer

rcv side reassemblessegments into messagespasses to app layer

more than one transportprotocol available to apps Internet TCP and UDP

applicationtransportnetworkdata linkphysical

applicationtransportnetworkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysicalnetwork

data linkphysical

logical end-end transport

3

Transport Layer 3-5

Transport vs network layer

network layer logicalcommunicationbetween hosts

transport layer logicalcommunicationbetween processes relies on enhances

network layer services

Household analogy12 kids sending letters

to 12 kids processes = kids app messages = letters

in envelopes hosts = houses transport protocol =

Ann and Bill network-layer protocol

= postal service

Transport Layer 3-6

Internet transport-layer protocols

reliable in-orderdelivery (TCP) congestion control flow control connection setup

unreliable unordereddelivery UDP no-frills extension of

ldquobest-effortrdquo IP services not available

delay guarantees bandwidth guarantees

applicationtransportnetworkdata linkphysical

applicationtransportnetworkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysicalnetwork

data linkphysical

logical end-end transport

4

Transport Layer 3-7

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-8

Multiplexingdemultiplexing

application

transport

network

link

physical

P1 application

transport

network

link

physical

application

transport

network

link

physical

P2P3 P4P1

host 1 host 2 host 3

= process= socket

delivering received segmentsto correct socket

Demultiplexing at rcv hostgathering data from multiplesockets enveloping data with header (later used for demultiplexing)

Multiplexing at send host

5

Transport Layer 3-9

How demultiplexing works host receives IP datagrams

each datagram has sourceIP address destination IPaddress

each datagram carries 1transport-layer segment

each segment has sourcedestination port number

host uses IP addresses amp portnumbers to direct segment toappropriate socket

source port dest port

32 bits

applicationdata

(message)

other header fields

TCPUDP segment format

Transport Layer 3-10

Connectionless demultiplexing

Create sockets with portnumbers

DatagramSocket mySocket1 = newDatagramSocket(12534)

DatagramSocket mySocket2 = newDatagramSocket(12535)

UDP socket identified bytwo-tuple

(dest IP address dest port number)

When host receives UDPsegment checks destination port

number in segment directs UDP segment to

socket with that portnumber

IP datagrams withdifferent source IPaddresses andor sourceport numbers directedto same socket

6

Transport Layer 3-11

Connectionless demux (cont)

DatagramSocket serverSocket = new DatagramSocket(6428)

ClientIPB

P2

client IP A

P1P1P3

serverIP C

SP 6428DP 9157

SP 9157DP 6428

SP 6428DP 5775

SP 5775DP 6428

SP provides ldquoreturn addressrdquo

Transport Layer 3-12

Connection-oriented demux

TCP socket identifiedby 4-tuple source IP address source port number dest IP address dest port number

recv host uses all fourvalues to directsegment to appropriatesocket

Server host may supportmany simultaneous TCPsockets each socket identified by

its own 4-tuple Web servers have

different sockets foreach connecting client non-persistent HTTP will

have different socket foreach request

7

Transport Layer 3-13

Connection-oriented demux(cont)

ClientIPB

P1

client IP A

P1P2P4

serverIP C

SP 9157DP 80

SP 9157DP 80

P5 P6 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

Transport Layer 3-14

Connection-oriented demuxThreaded Web Server - only 1process

ClientIPB

P1

client IP A

P1P2

serverIP C

SP 9157DP 80

SP 9157DP 80

P4 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

8

Transport Layer 3-15

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-16

UDP User Datagram Protocol [RFC 768]

ldquono frillsrdquo ldquobare bonesrdquoInternet transportprotocol

ldquobest effortrdquo service UDPsegments may be lost delivered out of order

to app connectionless

no handshaking betweenUDP sender receiver

each UDP segmenthandled independentlyof others

Why is there a UDP no connection

establishment (which canadd delay)

simple no connection stateat sender receiver

small segment header no congestion control UDP

can blast away as fast asdesired

9

Transport Layer 3-17

UDP more often used for streaming

multimedia apps loss tolerant rate sensitive

other UDP uses DNS SNMP

reliable transfer over UDPadd reliability atapplication layer application-specific

error recovery

source port dest port

32 bits

Applicationdata

(message)

UDP segment format

length checksumLength in

bytes of UDPsegmentincluding

header

Transport Layer 3-18

UDP checksum

Sender treat segment contents as

sequence of 16-bitintegers

checksum addition (1rsquoscomplement sum) ofsegment contents

sender puts checksumvalue into UDP checksumfield

Receiver compute checksum of

received segment check if computed checksum

equals checksum field value NO - error detected YES - no error detected

But maybe errorsnonetheless More laterhellip

Goal detect ldquoerrorsrdquo (eg flipped bits) intransmitted segment

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

3

Transport Layer 3-5

Transport vs network layer

network layer logicalcommunicationbetween hosts

transport layer logicalcommunicationbetween processes relies on enhances

network layer services

Household analogy12 kids sending letters

to 12 kids processes = kids app messages = letters

in envelopes hosts = houses transport protocol =

Ann and Bill network-layer protocol

= postal service

Transport Layer 3-6

Internet transport-layer protocols

reliable in-orderdelivery (TCP) congestion control flow control connection setup

unreliable unordereddelivery UDP no-frills extension of

ldquobest-effortrdquo IP services not available

delay guarantees bandwidth guarantees

applicationtransportnetworkdata linkphysical

applicationtransportnetworkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysical

networkdata linkphysicalnetwork

data linkphysical

logical end-end transport

4

Transport Layer 3-7

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-8

Multiplexingdemultiplexing

application

transport

network

link

physical

P1 application

transport

network

link

physical

application

transport

network

link

physical

P2P3 P4P1

host 1 host 2 host 3

= process= socket

delivering received segmentsto correct socket

Demultiplexing at rcv hostgathering data from multiplesockets enveloping data with header (later used for demultiplexing)

Multiplexing at send host

5

Transport Layer 3-9

How demultiplexing works host receives IP datagrams

each datagram has sourceIP address destination IPaddress

each datagram carries 1transport-layer segment

each segment has sourcedestination port number

host uses IP addresses amp portnumbers to direct segment toappropriate socket

source port dest port

32 bits

applicationdata

(message)

other header fields

TCPUDP segment format

Transport Layer 3-10

Connectionless demultiplexing

Create sockets with portnumbers

DatagramSocket mySocket1 = newDatagramSocket(12534)

DatagramSocket mySocket2 = newDatagramSocket(12535)

UDP socket identified bytwo-tuple

(dest IP address dest port number)

When host receives UDPsegment checks destination port

number in segment directs UDP segment to

socket with that portnumber

IP datagrams withdifferent source IPaddresses andor sourceport numbers directedto same socket

6

Transport Layer 3-11

Connectionless demux (cont)

DatagramSocket serverSocket = new DatagramSocket(6428)

ClientIPB

P2

client IP A

P1P1P3

serverIP C

SP 6428DP 9157

SP 9157DP 6428

SP 6428DP 5775

SP 5775DP 6428

SP provides ldquoreturn addressrdquo

Transport Layer 3-12

Connection-oriented demux

TCP socket identifiedby 4-tuple source IP address source port number dest IP address dest port number

recv host uses all fourvalues to directsegment to appropriatesocket

Server host may supportmany simultaneous TCPsockets each socket identified by

its own 4-tuple Web servers have

different sockets foreach connecting client non-persistent HTTP will

have different socket foreach request

7

Transport Layer 3-13

Connection-oriented demux(cont)

ClientIPB

P1

client IP A

P1P2P4

serverIP C

SP 9157DP 80

SP 9157DP 80

P5 P6 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

Transport Layer 3-14

Connection-oriented demuxThreaded Web Server - only 1process

ClientIPB

P1

client IP A

P1P2

serverIP C

SP 9157DP 80

SP 9157DP 80

P4 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

8

Transport Layer 3-15

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-16

UDP User Datagram Protocol [RFC 768]

ldquono frillsrdquo ldquobare bonesrdquoInternet transportprotocol

ldquobest effortrdquo service UDPsegments may be lost delivered out of order

to app connectionless

no handshaking betweenUDP sender receiver

each UDP segmenthandled independentlyof others

Why is there a UDP no connection

establishment (which canadd delay)

simple no connection stateat sender receiver

small segment header no congestion control UDP

can blast away as fast asdesired

9

Transport Layer 3-17

UDP more often used for streaming

multimedia apps loss tolerant rate sensitive

other UDP uses DNS SNMP

reliable transfer over UDPadd reliability atapplication layer application-specific

error recovery

source port dest port

32 bits

Applicationdata

(message)

UDP segment format

length checksumLength in

bytes of UDPsegmentincluding

header

Transport Layer 3-18

UDP checksum

Sender treat segment contents as

sequence of 16-bitintegers

checksum addition (1rsquoscomplement sum) ofsegment contents

sender puts checksumvalue into UDP checksumfield

Receiver compute checksum of

received segment check if computed checksum

equals checksum field value NO - error detected YES - no error detected

But maybe errorsnonetheless More laterhellip

Goal detect ldquoerrorsrdquo (eg flipped bits) intransmitted segment

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

4

Transport Layer 3-7

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-8

Multiplexingdemultiplexing

application

transport

network

link

physical

P1 application

transport

network

link

physical

application

transport

network

link

physical

P2P3 P4P1

host 1 host 2 host 3

= process= socket

delivering received segmentsto correct socket

Demultiplexing at rcv hostgathering data from multiplesockets enveloping data with header (later used for demultiplexing)

Multiplexing at send host

5

Transport Layer 3-9

How demultiplexing works host receives IP datagrams

each datagram has sourceIP address destination IPaddress

each datagram carries 1transport-layer segment

each segment has sourcedestination port number

host uses IP addresses amp portnumbers to direct segment toappropriate socket

source port dest port

32 bits

applicationdata

(message)

other header fields

TCPUDP segment format

Transport Layer 3-10

Connectionless demultiplexing

Create sockets with portnumbers

DatagramSocket mySocket1 = newDatagramSocket(12534)

DatagramSocket mySocket2 = newDatagramSocket(12535)

UDP socket identified bytwo-tuple

(dest IP address dest port number)

When host receives UDPsegment checks destination port

number in segment directs UDP segment to

socket with that portnumber

IP datagrams withdifferent source IPaddresses andor sourceport numbers directedto same socket

6

Transport Layer 3-11

Connectionless demux (cont)

DatagramSocket serverSocket = new DatagramSocket(6428)

ClientIPB

P2

client IP A

P1P1P3

serverIP C

SP 6428DP 9157

SP 9157DP 6428

SP 6428DP 5775

SP 5775DP 6428

SP provides ldquoreturn addressrdquo

Transport Layer 3-12

Connection-oriented demux

TCP socket identifiedby 4-tuple source IP address source port number dest IP address dest port number

recv host uses all fourvalues to directsegment to appropriatesocket

Server host may supportmany simultaneous TCPsockets each socket identified by

its own 4-tuple Web servers have

different sockets foreach connecting client non-persistent HTTP will

have different socket foreach request

7

Transport Layer 3-13

Connection-oriented demux(cont)

ClientIPB

P1

client IP A

P1P2P4

serverIP C

SP 9157DP 80

SP 9157DP 80

P5 P6 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

Transport Layer 3-14

Connection-oriented demuxThreaded Web Server - only 1process

ClientIPB

P1

client IP A

P1P2

serverIP C

SP 9157DP 80

SP 9157DP 80

P4 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

8

Transport Layer 3-15

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-16

UDP User Datagram Protocol [RFC 768]

ldquono frillsrdquo ldquobare bonesrdquoInternet transportprotocol

ldquobest effortrdquo service UDPsegments may be lost delivered out of order

to app connectionless

no handshaking betweenUDP sender receiver

each UDP segmenthandled independentlyof others

Why is there a UDP no connection

establishment (which canadd delay)

simple no connection stateat sender receiver

small segment header no congestion control UDP

can blast away as fast asdesired

9

Transport Layer 3-17

UDP more often used for streaming

multimedia apps loss tolerant rate sensitive

other UDP uses DNS SNMP

reliable transfer over UDPadd reliability atapplication layer application-specific

error recovery

source port dest port

32 bits

Applicationdata

(message)

UDP segment format

length checksumLength in

bytes of UDPsegmentincluding

header

Transport Layer 3-18

UDP checksum

Sender treat segment contents as

sequence of 16-bitintegers

checksum addition (1rsquoscomplement sum) ofsegment contents

sender puts checksumvalue into UDP checksumfield

Receiver compute checksum of

received segment check if computed checksum

equals checksum field value NO - error detected YES - no error detected

But maybe errorsnonetheless More laterhellip

Goal detect ldquoerrorsrdquo (eg flipped bits) intransmitted segment

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

5

Transport Layer 3-9

How demultiplexing works host receives IP datagrams

each datagram has sourceIP address destination IPaddress

each datagram carries 1transport-layer segment

each segment has sourcedestination port number

host uses IP addresses amp portnumbers to direct segment toappropriate socket

source port dest port

32 bits

applicationdata

(message)

other header fields

TCPUDP segment format

Transport Layer 3-10

Connectionless demultiplexing

Create sockets with portnumbers

DatagramSocket mySocket1 = newDatagramSocket(12534)

DatagramSocket mySocket2 = newDatagramSocket(12535)

UDP socket identified bytwo-tuple

(dest IP address dest port number)

When host receives UDPsegment checks destination port

number in segment directs UDP segment to

socket with that portnumber

IP datagrams withdifferent source IPaddresses andor sourceport numbers directedto same socket

6

Transport Layer 3-11

Connectionless demux (cont)

DatagramSocket serverSocket = new DatagramSocket(6428)

ClientIPB

P2

client IP A

P1P1P3

serverIP C

SP 6428DP 9157

SP 9157DP 6428

SP 6428DP 5775

SP 5775DP 6428

SP provides ldquoreturn addressrdquo

Transport Layer 3-12

Connection-oriented demux

TCP socket identifiedby 4-tuple source IP address source port number dest IP address dest port number

recv host uses all fourvalues to directsegment to appropriatesocket

Server host may supportmany simultaneous TCPsockets each socket identified by

its own 4-tuple Web servers have

different sockets foreach connecting client non-persistent HTTP will

have different socket foreach request

7

Transport Layer 3-13

Connection-oriented demux(cont)

ClientIPB

P1

client IP A

P1P2P4

serverIP C

SP 9157DP 80

SP 9157DP 80

P5 P6 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

Transport Layer 3-14

Connection-oriented demuxThreaded Web Server - only 1process

ClientIPB

P1

client IP A

P1P2

serverIP C

SP 9157DP 80

SP 9157DP 80

P4 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

8

Transport Layer 3-15

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-16

UDP User Datagram Protocol [RFC 768]

ldquono frillsrdquo ldquobare bonesrdquoInternet transportprotocol

ldquobest effortrdquo service UDPsegments may be lost delivered out of order

to app connectionless

no handshaking betweenUDP sender receiver

each UDP segmenthandled independentlyof others

Why is there a UDP no connection

establishment (which canadd delay)

simple no connection stateat sender receiver

small segment header no congestion control UDP

can blast away as fast asdesired

9

Transport Layer 3-17

UDP more often used for streaming

multimedia apps loss tolerant rate sensitive

other UDP uses DNS SNMP

reliable transfer over UDPadd reliability atapplication layer application-specific

error recovery

source port dest port

32 bits

Applicationdata

(message)

UDP segment format

length checksumLength in

bytes of UDPsegmentincluding

header

Transport Layer 3-18

UDP checksum

Sender treat segment contents as

sequence of 16-bitintegers

checksum addition (1rsquoscomplement sum) ofsegment contents

sender puts checksumvalue into UDP checksumfield

Receiver compute checksum of

received segment check if computed checksum

equals checksum field value NO - error detected YES - no error detected

But maybe errorsnonetheless More laterhellip

Goal detect ldquoerrorsrdquo (eg flipped bits) intransmitted segment

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

6

Transport Layer 3-11

Connectionless demux (cont)

DatagramSocket serverSocket = new DatagramSocket(6428)

ClientIPB

P2

client IP A

P1P1P3

serverIP C

SP 6428DP 9157

SP 9157DP 6428

SP 6428DP 5775

SP 5775DP 6428

SP provides ldquoreturn addressrdquo

Transport Layer 3-12

Connection-oriented demux

TCP socket identifiedby 4-tuple source IP address source port number dest IP address dest port number

recv host uses all fourvalues to directsegment to appropriatesocket

Server host may supportmany simultaneous TCPsockets each socket identified by

its own 4-tuple Web servers have

different sockets foreach connecting client non-persistent HTTP will

have different socket foreach request

7

Transport Layer 3-13

Connection-oriented demux(cont)

ClientIPB

P1

client IP A

P1P2P4

serverIP C

SP 9157DP 80

SP 9157DP 80

P5 P6 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

Transport Layer 3-14

Connection-oriented demuxThreaded Web Server - only 1process

ClientIPB

P1

client IP A

P1P2

serverIP C

SP 9157DP 80

SP 9157DP 80

P4 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

8

Transport Layer 3-15

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-16

UDP User Datagram Protocol [RFC 768]

ldquono frillsrdquo ldquobare bonesrdquoInternet transportprotocol

ldquobest effortrdquo service UDPsegments may be lost delivered out of order

to app connectionless

no handshaking betweenUDP sender receiver

each UDP segmenthandled independentlyof others

Why is there a UDP no connection

establishment (which canadd delay)

simple no connection stateat sender receiver

small segment header no congestion control UDP

can blast away as fast asdesired

9

Transport Layer 3-17

UDP more often used for streaming

multimedia apps loss tolerant rate sensitive

other UDP uses DNS SNMP

reliable transfer over UDPadd reliability atapplication layer application-specific

error recovery

source port dest port

32 bits

Applicationdata

(message)

UDP segment format

length checksumLength in

bytes of UDPsegmentincluding

header

Transport Layer 3-18

UDP checksum

Sender treat segment contents as

sequence of 16-bitintegers

checksum addition (1rsquoscomplement sum) ofsegment contents

sender puts checksumvalue into UDP checksumfield

Receiver compute checksum of

received segment check if computed checksum

equals checksum field value NO - error detected YES - no error detected

But maybe errorsnonetheless More laterhellip

Goal detect ldquoerrorsrdquo (eg flipped bits) intransmitted segment

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

7

Transport Layer 3-13

Connection-oriented demux(cont)

ClientIPB

P1

client IP A

P1P2P4

serverIP C

SP 9157DP 80

SP 9157DP 80

P5 P6 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

Transport Layer 3-14

Connection-oriented demuxThreaded Web Server - only 1process

ClientIPB

P1

client IP A

P1P2

serverIP C

SP 9157DP 80

SP 9157DP 80

P4 P3

D-IPCS-IP AD-IPC

S-IP B

SP 5775DP 80

D-IPCS-IP B

8

Transport Layer 3-15

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-16

UDP User Datagram Protocol [RFC 768]

ldquono frillsrdquo ldquobare bonesrdquoInternet transportprotocol

ldquobest effortrdquo service UDPsegments may be lost delivered out of order

to app connectionless

no handshaking betweenUDP sender receiver

each UDP segmenthandled independentlyof others

Why is there a UDP no connection

establishment (which canadd delay)

simple no connection stateat sender receiver

small segment header no congestion control UDP

can blast away as fast asdesired

9

Transport Layer 3-17

UDP more often used for streaming

multimedia apps loss tolerant rate sensitive

other UDP uses DNS SNMP

reliable transfer over UDPadd reliability atapplication layer application-specific

error recovery

source port dest port

32 bits

Applicationdata

(message)

UDP segment format

length checksumLength in

bytes of UDPsegmentincluding

header

Transport Layer 3-18

UDP checksum

Sender treat segment contents as

sequence of 16-bitintegers

checksum addition (1rsquoscomplement sum) ofsegment contents

sender puts checksumvalue into UDP checksumfield

Receiver compute checksum of

received segment check if computed checksum

equals checksum field value NO - error detected YES - no error detected

But maybe errorsnonetheless More laterhellip

Goal detect ldquoerrorsrdquo (eg flipped bits) intransmitted segment

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

8

Transport Layer 3-15

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-16

UDP User Datagram Protocol [RFC 768]

ldquono frillsrdquo ldquobare bonesrdquoInternet transportprotocol

ldquobest effortrdquo service UDPsegments may be lost delivered out of order

to app connectionless

no handshaking betweenUDP sender receiver

each UDP segmenthandled independentlyof others

Why is there a UDP no connection

establishment (which canadd delay)

simple no connection stateat sender receiver

small segment header no congestion control UDP

can blast away as fast asdesired

9

Transport Layer 3-17

UDP more often used for streaming

multimedia apps loss tolerant rate sensitive

other UDP uses DNS SNMP

reliable transfer over UDPadd reliability atapplication layer application-specific

error recovery

source port dest port

32 bits

Applicationdata

(message)

UDP segment format

length checksumLength in

bytes of UDPsegmentincluding

header

Transport Layer 3-18

UDP checksum

Sender treat segment contents as

sequence of 16-bitintegers

checksum addition (1rsquoscomplement sum) ofsegment contents

sender puts checksumvalue into UDP checksumfield

Receiver compute checksum of

received segment check if computed checksum

equals checksum field value NO - error detected YES - no error detected

But maybe errorsnonetheless More laterhellip

Goal detect ldquoerrorsrdquo (eg flipped bits) intransmitted segment

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

9

Transport Layer 3-17

UDP more often used for streaming

multimedia apps loss tolerant rate sensitive

other UDP uses DNS SNMP

reliable transfer over UDPadd reliability atapplication layer application-specific

error recovery

source port dest port

32 bits

Applicationdata

(message)

UDP segment format

length checksumLength in

bytes of UDPsegmentincluding

header

Transport Layer 3-18

UDP checksum

Sender treat segment contents as

sequence of 16-bitintegers

checksum addition (1rsquoscomplement sum) ofsegment contents

sender puts checksumvalue into UDP checksumfield

Receiver compute checksum of

received segment check if computed checksum

equals checksum field value NO - error detected YES - no error detected

But maybe errorsnonetheless More laterhellip

Goal detect ldquoerrorsrdquo (eg flipped bits) intransmitted segment

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

10

Transport Layer 3-19

Internet Checksum Example Note

When adding numbers a carryout from themost significant bit needs to be added to theresult

Example add two 16-bit integers

1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 1 01 1 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 0 1 1

1 1 0 1 1 1 0 1 1 1 0 1 1 1 1 0 01 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 1

wraparound

sumchecksum

Transport Layer 3-20

Internet Checksum Why does UDP provide a checksum at all Many

link-layer protocols also provide error-checking

Answero No guarantee that all the links between source and

destination provide error-checkingo Even if segments are correctly transferred bit-

errors could be introduced when a segment isstored in routerrsquos memory

UDP must provide error detection on an end-to-end basisApplication of the end-to-end principleCertain functionality must be implemented on an end-end basis

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

11

Transport Layer 3-21

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-22

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

12

Transport Layer 3-23

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

Transport Layer 3-24

Principles of Reliable data transfer important in app transport link layers top-10 list of important networking topics

characteristics of unreliable channel will determinecomplexity of reliable data transfer protocol (rdt)

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

13

Transport Layer 3-25

Reliable data transfer getting started

sendside

receiveside

rdt_send() called from above(eg by app) Passed data to

deliver to receiver upper layer

udt_send() called by rdtto transfer packet over

unreliable channel to receiver

rdt_rcv() called when packetarrives on rcv-side of channel

deliver_data() called byrdt to deliver data to upper

Transport Layer 3-26

Reliable data transfer getting startedWersquoll incrementally develop sender receiver sides of

reliable data transfer protocol (rdt) consider only unidirectional data transfer

but control info will flow on both directions use finite state machines (FSM) to specify

sender receiver

state1

state2

event causing state transitionactions taken on state transition

state when in thisldquostaterdquo next state

uniquely determinedby next event

eventactions

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

14

Transport Layer 3-27

Rdt10 reliable transfer over a reliable channel

underlying channel perfectly reliable no bit errors no loss of packets

separate FSMs for sender receiver sender sends data into underlying channel receiver read data from underlying channel

Wait forcall fromabove packet = make_pkt(data)

udt_send(packet)

rdt_send(data)extract (packetdata)deliver_data(data)

Wait forcall from

below

rdt_rcv(packet)

sender receiver

Transport Layer 3-28

Rdt20 channel with bit errors underlying channel may flip bits in packet

checksum to detect bit errors the question how to recover from errors

acknowledgements (ACKs) receiver explicitly tells senderthat pkt received OK

negative acknowledgements (NAKs) receiver explicitlytells sender that pkt had errors

sender retransmits pkt on receipt of NAK known as ARQ (Automatic Repeat reQuest) protocols

new mechanisms in rdt20 (beyond rdt10) error detection receiver feedback control msgs (ACKNAK) rcvr-gtsender

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

15

Transport Layer 3-29

rdt20 FSM specification

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

belowsender

receiverrdt_send(data)

Λ

Λ wait

Transport Layer 3-30

rdt20 operation with no errors

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

16

Transport Layer 3-31

rdt20 error scenario

Wait forcall fromabove

snkpkt = make_pkt(data checksum)udt_send(sndpkt)

extract(rcvpktdata)deliver_data(data)udt_send(ACK)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

rdt_rcv(rcvpkt) ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp isNAK(rcvpkt)

udt_send(NAK)

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Wait forACK or

NAK

Wait forcall from

below

rdt_send(data)

Λ

Transport Layer 3-32

rdt20 has a fatal flaw

What happens ifACKNAK corrupted

sender doesnrsquot know whathappened at receiver

canrsquot just retransmitpossible duplicate

Handling duplicates sender retransmits current

pkt if ACKNAK garbled sender adds sequence

number to each pkt receiver discards (doesnrsquot

deliver up) duplicate pkt

Sender sends one packet then waits for receiver response

stop and wait

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

17

Transport Layer 3-33

rdt21 sender handles garbled ACKNAKs

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

Wait forACK orNAK 0 udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isNAK(rcvpkt) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt)

Wait for call 1 from

above

Wait forACK orNAK 1

ΛΛ

Transport Layer 3-34

rdt21 receiver handles garbled ACKNAKs

Wait for0 frombelow

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq0(rcvpkt)

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

Wait for1 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq0(rcvpkt)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp not corrupt(rcvpkt) ampamp has_seq1(rcvpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt)

sndpkt = make_pkt(ACK chksum)udt_send(sndpkt)

sndpkt = make_pkt(NAK chksum)udt_send(sndpkt)

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

18

Transport Layer 3-35

rdt21 discussion

Sender seq added to pkt two seq rsquos (01) will

suffice Why must check if received

ACKNAK corrupted twice as many states

state must ldquorememberrdquowhether ldquocurrentrdquo pkthas 0 or 1 seq

Receiver must check if received

packet is duplicate state indicates whether

0 or 1 is expected pktseq

note receiver can notknow if its lastACKNAK received OKat sender

Transport Layer 3-36

rdt22 a NAK-free protocol

same functionality as rdt21 using ACKs only instead of NAK receiver sends ACK for last pkt

received OK receiver must explicitly include seq of pkt being ACKed

duplicate ACK at sender results in same action asNAK retransmit current pkt

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

19

Transport Layer 3-37

rdt22 sender receiver fragments

Wait forcall 0 from

above

sndpkt = make_pkt(0 data checksum)udt_send(sndpkt)

rdt_send(data)

udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) || isACK(rcvpkt1) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

Wait forACK

0sender FSM

fragment

Wait for0 frombelow

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt) ampamp has_seq1(rcvpkt)extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(ACK1 chksum)udt_send(sndpkt)

rdt_rcv(rcvpkt) ampamp (corrupt(rcvpkt) || has_seq1(rcvpkt))

udt_send(sndpkt)receiver FSM

fragment

Λ

Transport Layer 3-38

rdt30 channels with errors and loss

New assumptionunderlying channel canalso lose packets (dataor ACKs) checksum seq ACKs

retransmissions will beof help but not enough

Approach sender waitsldquoreasonablerdquo amount oftime for ACK

retransmits if no ACKreceived in this time

if pkt (or ACK) just delayed(not lost) retransmission will be

duplicate but use of seqrsquos already handles this

receiver must specify seq of pkt being ACKed

requires countdown timer

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

20

Transport Layer 3-39

rdt30 sendersndpkt = make_pkt(0 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

Waitfor

ACK0

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt1) )

Wait forcall 1 from

above

sndpkt = make_pkt(1 data checksum)udt_send(sndpkt)start_timer

rdt_send(data)

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt0)

rdt_rcv(rcvpkt) ampamp( corrupt(rcvpkt) ||isACK(rcvpkt0) )

rdt_rcv(rcvpkt)ampamp notcorrupt(rcvpkt)ampamp isACK(rcvpkt1)

stop_timerstop_timer

udt_send(sndpkt)start_timer

timeout

udt_send(sndpkt)start_timer

timeout

rdt_rcv(rcvpkt)

Wait forcall 0from

above

Waitfor

ACK1

Λrdt_rcv(rcvpkt)

ΛΛ

Λ

Transport Layer 3-40

rdt30 in action

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

21

Transport Layer 3-41

rdt30 in action

Transport Layer 3-42

Performance of rdt30

rdt30 works but performance stinks example 1 Gbps link 15 ms e-e prop delay 1KB packet

Ttransmit = 8kbpkt109 bsec = 8 microsec

U sender utilization ndash fraction of time sender busy sending

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

L (packet length in bits)R (transmission rate bps) =

1KB pkt every 30 msec -gt 33kBsec thruput over 1 Gbps link network protocol limits use of physical resources

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

22

Transport Layer 3-43

rdt30 stop-and-wait operation

first packet bit transmitted t = 0

sender receiver

RTT

last packet bit transmitted t = L R

first packet bit arriveslast packet bit arrives sendACK

ACK arrives send nextpacket t = RTT + L R

U sender

= 008

30008 = 000027

microsec

onds

L R

RTT + L R =

Transport Layer 3-44

Pipelined protocolsPipelining sender allows multiple ldquoin-flightrdquo yet-to-

be-acknowledged pkts range of sequence numbers must be increased buffering at sender andor receiver

Two generic forms of pipelined protocols go-Back-Nselective repeat

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

23

Transport Layer 3-45

Pipelining increased utilization

first packet bit transmitted t = 0

sender receiver

RTT

last bit transmitted t = L R

first packet bit arriveslast packet bit arrives send ACK

ACK arrives send nextpacket t = RTT + L R

last bit of 2nd packet arrives send ACKlast bit of 3rd packet arrives send ACK

U sender

= 024

30008 = 00008

microsecon

ds

3 L R

RTT + L R =

Increase utilizationby a factor of 3

Transport Layer 3-46

Go-Back-NSender k-bit seq in pkt header ldquowindowrdquo of up to N consecutive unackrsquoed pkts allowed

ACK(n) ACKs all pkts up to including seq n - ldquocumulative ACKrdquo may receive duplicate ACKs (see receiver)

timer for each in-flight pkt timeout(n) retransmit pkt n and all higher seq pkts in window

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

24

Transport Layer 3-47

GBN sender extended FSM

Wait start_timerudt_send(sndpkt[base])udt_send(sndpkt[base+1])hellipudt_send(sndpkt[nextseqnum-1])

timeout

rdt_send(data)

if (nextseqnum lt base+N) sndpkt[nextseqnum] = make_pkt(nextseqnumdatachksum) udt_send(sndpkt[nextseqnum]) if (base == nextseqnum) start_timer nextseqnum++ else refuse_data(data)

base = getacknum(rcvpkt)+1If (base == nextseqnum) stop_timer else start_timer

rdt_rcv(rcvpkt) ampamp notcorrupt(rcvpkt)

base=1nextseqnum=1

rdt_rcv(rcvpkt) ampamp corrupt(rcvpkt)

Λ

Transport Layer 3-48

GBN receiver extended FSM

ACK-only always send ACK for correctly-received pktwith highest in-order seq may generate duplicate ACKs need only remember expectedseqnum

out-of-order pkt discard (donrsquot buffer) -gt no receiver buffering Re-ACK pkt with highest in-order seq

Wait

udt_send(sndpkt)default

rdt_rcv(rcvpkt) ampamp notcurrupt(rcvpkt) ampamp hasseqnum(rcvpktexpectedseqnum)

extract(rcvpktdata)deliver_data(data)sndpkt = make_pkt(expectedseqnumACKchksum)udt_send(sndpkt)expectedseqnum++

expectedseqnum=1sndpkt = make_pkt(expectedseqnumACKchksum)

Λ

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

25

Transport Layer 3-49

GBN inaction

Transport Layer 3-50

Selective Repeat

receiver individually acknowledges all correctlyreceived pkts buffers pkts as needed for eventual in-order delivery

to upper layer sender only resends pkts for which ACK not

received sender timer for each unACKed pkt

sender window N consecutive seq rsquos again limits seq s of sent unACKed pkts

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

26

Transport Layer 3-51

Selective repeat sender receiver windows

Transport Layer 3-52

Selective repeat

data from above if next available seq in

window send pkttimeout(n) resend pkt n restart timerACK(n) in [sendbasesendbase+N]

mark pkt n as received if n smallest unACKed pkt

advance window base tonext unACKed seq

senderpkt n in [rcvbase rcvbase+N-1]

send ACK(n) out-of-order buffer in-order deliver (also

deliver buffered in-orderpkts) advance window tonext not-yet-received pkt

pkt n in [rcvbase-Nrcvbase-1]

ACK(n)otherwise ignore

receiver

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

27

Transport Layer 3-53

Selective repeat in action

Transport Layer 3-54

Selective repeat dilemmaExample seq rsquos 0 1 2 3 window size=3

receiver sees nodifference in twoscenarios

incorrectly passesduplicate data as newin (a)

Q what relationshipbetween seq sizeand window size

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

28

Transport Layer 3-55

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-56

TCP Overview RFCs 793 1122 1323 2018 2581

full duplex data bi-directional data flow

in same connection MSS maximum segment

size connection-oriented

handshaking (exchangeof control msgs) initrsquossender receiver statebefore data exchange

flow controlled sender will not

overwhelm receiver

point-to-point one sender one receiver

reliable in-order bytesteam no ldquomessage boundariesrdquo

pipelined TCP congestion and flow

control set window size send amp receive buffers

socket

door

TCP

send buffer

TCP

receive buffer

socket

door

segment

application

writes data

application

reads data

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

29

Transport Layer 3-57

TCP segment structure

source port dest port

32 bits

applicationdata

(variable length)

sequence numberacknowledgement number

Receive windowUrg data pnterchecksum

FSRPAUheadlen

notused

Options (variable length)

URG urgent data (generally not used)

ACK ACK valid

PSH push data now(generally not used)

RST SYN FINconnection estab(setup teardown

commands)

bytes rcvr willingto accept

countingby bytes of data(not segments)

Internetchecksum

(as in UDP)

Transport Layer 3-58

TCP seq rsquos and ACKsSeq rsquos

byte streamldquonumberrdquo of firstbyte in segmentrsquosdata

ACKs seq of next byte

expected fromother side

cumulative ACKQ how receiver handles

out-of-order segments A TCP spec doesnrsquot

say - up toimplementor

Host A Host B

Seq=42 ACK=79 data = lsquoCrsquo

Seq=79 ACK=43 data = lsquoCrsquo

Seq=43 ACK=80

Usertypes

lsquoCrsquo

host ACKsreceipt

of echoedlsquoCrsquo

host ACKsreceipt oflsquoCrsquo echoes

back lsquoCrsquo

timesimple telnet scenario

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

30

Transport Layer 3-59

TCP Round Trip Time and Timeout

Q how to set TCPtimeout value

longer than RTT but RTT varies

too short prematuretimeout unnecessary

retransmissions too long slow reaction

to segment loss

Q how to estimate RTT SampleRTT measured time from

segment transmission until ACKreceipt ignore retransmissions

SampleRTT will vary wantestimated RTT ldquosmootherrdquo average several recent

measurements not justcurrent SampleRTT

Transport Layer 3-60

TCP Round Trip Time and Timeout

EstimatedRTT = (1- α)EstimatedRTT + αSampleRTT

Exponential weighted moving average influence of past sample decreases exponentially fast typical value α = 0125

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

31

Transport Layer 3-61

Example RTT estimationRTT gaiacsumassedu to fantasiaeurecomfr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RT

T

(m

illise

co

nd

s)

SampleRTT Estimated RTT

Transport Layer 3-62

TCP Round Trip Time and Timeout

Setting the timeout EstimtedRTT plus ldquosafety marginrdquo

large variation in EstimatedRTT -gt larger safety margin first estimate of how much SampleRTT deviates from

EstimatedRTT

TimeoutInterval = EstimatedRTT + 4DevRTT

DevRTT = (1-β)DevRTT + β|SampleRTT-EstimatedRTT|

(typically β = 025)

Then set timeout interval

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

32

Transport Layer 3-63

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-64

TCP reliable data transfer

TCP creates rdtservice on top of IPrsquosunreliable service

Pipelined segments Cumulative acks TCP uses single

retransmission timer

Retransmissions aretriggered by timeout events duplicate acks

Initially considersimplified TCP sender ignore duplicate acks ignore flow control

congestion control

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

33

Transport Layer 3-65

TCP sender eventsdata rcvd from app Create segment with

seq seq is byte-stream

number of first databyte in segment

start timer if notalready running (thinkof timer as for oldestunacked segment)

expiration intervalTimeOutInterval

timeout retransmit segment

that caused timeout restart timer Ack rcvd If acknowledges

previously unackedsegments update what is known to

be acked start timer if there are

outstanding segments

Transport Layer 3-66

TCP retransmission scenariosHost A

Seq=100 20 bytes data

ACK=100

timepremature timeout

Host B

Seq=92 8 bytes data

ACK=120

Seq=92 8 bytes data

Seq=

92 t

imeo

ut

ACK=120

Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

lost ACK scenario

Host B

X

Seq=92 8 bytes data

ACK=100

time

Seq=

92 t

imeo

ut

SendBase= 100

SendBase= 120

SendBase= 120

Sendbase= 100

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

34

Transport Layer 3-67

TCP retransmission scenarios (more)Host A

Seq=92 8 bytes data

ACK=100

loss

tim

eout

Cumulative ACK scenario

Host B

X

Seq=100 20 bytes data

ACK=120

time

SendBase= 120

Transport Layer 3-68

TCP ACK generation [RFC 1122 RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq All data up toexpected seq already ACKed

Arrival of in-order segment withexpected seq One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK Wait up to 500msfor next segment If no next segmentsend ACK

Immediately send single cumulative ACK ACKing both in-order segments

Immediately send duplicate ACK indicating seq of next expected byte

Immediate send ACK provided thatsegment startsat lower end of gap

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

35

Transport Layer 3-69

Fast Retransmit

Time-out period oftenrelatively long long delay before

resending lost packet Detect lost segments

via duplicate ACKs Sender often sends

many segments back-to-back

If segment is lostthere will likely be manyduplicate ACKs

If sender receives 3ACKs for the samedata it supposes thatsegment after ACKeddata was lost fast retransmit resend

segment before timerexpires

Transport Layer 3-70

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

36

Transport Layer 3-71

TCP Flow Control

receive side of TCPconnection has areceive buffer

speed-matchingservice matching thesend rate to thereceiving apprsquos drainrate

app process may beslow at reading frombuffer

sender wonrsquot overflowreceiverrsquos buffer by

transmitting toomuch

too fast

flow control

Transport Layer 3-72

TCP Flow control how it works

(Suppose TCP receiverdiscards out-of-ordersegments)

spare room in buffer= RcvWindow= RcvBuffer-[LastByteRcvd -

LastByteRead]

Rcvr advertises spareroom by including valueof RcvWindow insegments

Sender limits unACKeddata to RcvWindow guarantees receive

buffer doesnrsquot overflow

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

37

Transport Layer 3-73

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

Transport Layer 3-74

TCP Connection ManagementRecall TCP sender receiver

establish ldquoconnectionrdquobefore exchanging datasegments

initialize TCP variables seq s buffers flow control

info (eg RcvWindow) client connection initiator Socket clientSocket = new

Socket(hostnameportnumber)

server contacted by client Socket connectionSocket =

welcomeSocketaccept()

Three way handshakeStep 1 client host sends TCP

SYN segment to server specifies initial seq no data

Step 2 server host receivesSYN replies with SYNACKsegment server allocates buffers specifies server initial seq

Step 3 client receives SYNACK

replies with ACK segmentwhich may contain data

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

38

Transport Layer 3-75

TCP Connection Management (cont)

Closing a connectionclient closes socket

clientSocketclose()

Step 1 client end systemsends TCP FIN controlsegment to server

Step 2 server receivesFIN replies with ACKCloses connection sendsFIN

client

FIN

server

ACK

ACK

FIN

close

close

closedti

med

wai

t

Transport Layer 3-76

TCP Connection Management (cont)

Step 3 client receives FINreplies with ACK

Enters ldquotimed waitrdquo -will respond with ACKto received FINs

Step 4 server receivesACK Connection closed

Note with smallmodification can handlesimultaneous FINs

client

FIN

server

ACK

ACK

FIN

closing

closing

closed

tim

ed w

ait

closed

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

39

Transport Layer 3-77

TCP Connection Management (cont)

TCP clientlifecycle

TCP serverlifecycle

Transport Layer 3-78

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

40

Transport Layer 3-79

Principles of Congestion Control

Congestion informally ldquotoo many sources sending too much

data too fast for network to handlerdquo different from flow control manifestations

lost packets (buffer overflow at routers) long delays (queueing in router buffers)

a top-10 problem

Transport Layer 3-80

Causescosts of congestion scenario 1

two senders tworeceivers

one routerinfinite buffers

no retransmission

large delayswhen congested

maximumachievablethroughput

unlimited sharedoutput link buffers

Host Aλin original data

Host B

λout

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

41

Transport Layer 3-81

Causescosts of congestion scenario 2

one router finite buffers sender retransmission of lost packet

finite shared outputlink buffers

Host A λin originaldata

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-82

Causescosts of congestion scenario 2 always (goodput) ldquoperfectrdquo retransmission only when loss

retransmission of delayed (not lost) packet makes larger(than perfect case) for same

λin

λout=

λin

λoutgtλ

inλ

out

ldquocostsrdquo of congestion more work (retrans) for given ldquogoodputrdquo unneeded retransmissions link carries multiple copies of pkt

R2

R2λin

λ out

b

R2

R2λin

λ out

a

R2

R2λin

λ out

c

R4

R3

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

42

Transport Layer 3-83

Causescosts of congestion scenario 3 four senders multihop paths timeoutretransmit

λin

Q what happens asand increase λ

in

finite shared outputlink buffers

Host Aλin original data

Host B

λout

λin original data plusretransmitted data

Transport Layer 3-84

Causescosts of congestion scenario 3

Another ldquocostrdquo of congestion when packet dropped any ldquoupstream transmission

capacity used for that packet was wasted

HostA

HostB

λout

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

43

Transport Layer 3-85

Approaches towards congestion control

End-end congestioncontrol

no explicit feedback fromnetwork

congestion inferred fromend-system observed lossdelay

approach taken by TCP

Network-assistedcongestion control

routers provide feedbackto end systems single bit indicating

congestion (SNADECbit TCPIP ECNATM)

explicit rate sendershould send at

Two broad approaches towards congestion control

Transport Layer 3-86

Case study ATM ABR congestion control

ABR available bit rate ldquoelastic servicerdquo if senderrsquos path

ldquounderloadedrdquo sender should use

available bandwidth if senderrsquos path

congested sender throttled to

minimum guaranteedrate

RM (resource management)cells

sent by sender interspersedwith data cells

bits in RM cell set by switches(ldquonetwork-assistedrdquo) NI bit no increase in rate

(mild congestion) CI bit congestion

indication RM cells returned to sender by

receiver with bits intact

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

44

Transport Layer 3-87

Case study ATM ABR congestion control

two-byte ER (explicit rate) field in RM cell congested switch may lower ER value in cell senderrsquo send rate thus maximum supportable rate on path

EFCI bit in data cells set to 1 in congested switch if data cell preceding RM cell has EFCI set sender sets CI

bit in returned RM cell

Transport Layer 3-88

Chapter 3 outline

31 Transport-layerservices

32 Multiplexing anddemultiplexing

33 Connectionlesstransport UDP

34 Principles ofreliable data transfer

35 Connection-orientedtransport TCP segment structure reliable data transfer flow control connection management

36 Principles ofcongestion control

37 TCP congestioncontrol

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

45

Transport Layer 3-89

TCP congestion control additive increasemultiplicative decrease

8 Kbytes

16 Kbytes

24 Kbytes

time

congestion

window

Approach increase transmission rate (window size)probing for usable bandwidth until loss occurs additive increase increase CongWin by 1 MSS

every RTT until loss detected multiplicative decrease cut CongWin in half after

loss

timecong

estio

n w

indo

w s

ize

Saw toothbehavior probing

for bandwidth

Transport Layer 3-90

TCP Congestion Control details

sender limits transmission LastByteSent-LastByteAcked le CongWin

Roughly

CongWin is dynamic functionof perceived networkcongestion

How does senderperceive congestion

loss event = timeout or3 duplicate acks

TCP sender reducesrate (CongWin) afterloss event

three mechanisms AIMD slow start conservative after

timeout events

rate = CongWin

RTT Bytessec

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

46

Transport Layer 3-91

TCP Slow Start

When connection beginsCongWin = 1 MSS Example MSS = 500

bytes amp RTT = 200 msec initial rate = 20 kbps

available bandwidth maybe gtgt MSSRTT desirable to quickly ramp

up to respectable rate

When connection beginsincrease rateexponentially fast untilfirst loss event

Transport Layer 3-92

TCP Slow Start (more)

When connectionbegins increase rateexponentially untilfirst loss event double CongWin every

RTT done by incrementingCongWin for every ACKreceived

Summary initial rateis slow but ramps upexponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

47

Transport Layer 3-93

RefinementQ When should the

exponentialincrease switch tolinear

A When CongWingets to 12 of itsvalue beforetimeout

Implementation Variable Threshold At loss event Threshold is

set to 12 of CongWin justbefore loss event

Transport Layer 3-94

Refinement inferring loss

After 3 dup ACKs CongWin is cut in half window then grows

linearly But after timeout event

CongWin instead set to1 MSS

window then growsexponentially

to a threshold thengrows linearly

3 dup ACKs indicatesnetwork capable ofdelivering somesegments timeout indicates aldquomore alarmingrdquocongestion scenario

Philosophy

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

48

Transport Layer 3-95

Summary TCP Congestion Control

When CongWin is below Threshold sender inslow-start phase window grows exponentially

When CongWin is above Threshold sender is incongestion-avoidance phase window grows linearly

When a triple duplicate ACK occurs Thresholdset to CongWin2 and CongWin set toThreshold

When timeout occurs Threshold set toCongWin2 and CongWin is set to 1 MSS

Transport Layer 3-96

TCP sender congestion control

SS or CA

SS or CA

SS or CA

CongestionAvoidance(CA)

Slow Start(SS)

State

CongWin and Thresholdnot changed

Increment duplicate ACK countfor segment being acked

DuplicateACK

Enter slow startThreshold = CongWin2CongWin = 1 MSSSet state to ldquoSlow Startrdquo

Timeout

Fast recoveryimplementing multiplicativedecrease CongWin will notdrop below 1 MSS

Threshold = CongWin2CongWin = ThresholdSet state to ldquoCongestionAvoidancerdquo

Loss eventdetected bytripleduplicateACK

Additive increase resultingin increase of CongWin by1 MSS every RTT

CongWin = CongWin+MSS (MSSCongWin)

ACK receiptfor previouslyunackeddata

Resulting in a doubling ofCongWin every RTT

CongWin = CongWin + MSSIf (CongWin gt Threshold) set state to ldquoCongestionAvoidancerdquo

ACK receiptfor previouslyunackeddata

CommentaryTCP Sender ActionEvent

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

49

Transport Layer 3-97

TCP throughput

Whatrsquos the average throughout of TCP as afunction of window size and RTT Ignore slow start

Let W be the window size when loss occurs When window is W throughput is WRTT Just after loss window drops to W2

throughput to W2RTT Average throughout 75 WRTT

Transport Layer 3-98

TCP Futures TCP over ldquolong fat pipesrdquo

Example 1500 byte segments 100ms RTT want 10Gbps throughput

Requires window size W = 83333 in-flightsegments

Throughput in terms of loss rate

L = 210-10 Wow New versions of TCP for high-speed needed

LRTT

MSS221

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

50

Transport Layer 3-99

Fairness goal if K TCP sessions share samebottleneck link of bandwidth R each should haveaverage rate of RK

TCP connection 1

bottleneckrouter

capacity R

TCP connection 2

TCP Fairness

Transport Layer 3-100

Why is TCP fairTwo competing sessions Additive increase gives slope of 1 as throughout increases multiplicative decrease decreases throughput proportionally

R

R

equal bandwidth share

Connection 1 throughput

Conn

ecti

on 2

thr

ough

put

congestion avoidance additive increaseloss decrease window by factor of 2

congestion avoidance additive increaseloss decrease window by factor of 2

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

51

Transport Layer 3-101

Fairness (more)Fairness and UDP Multimedia apps often

do not use TCP do not want rate

throttled by congestioncontrol

Instead use UDP pump audiovideo at

constant rate toleratepacket loss

Research area TCPfriendly

Fairness and parallel TCPconnections

nothing prevents app fromopening parallelconnections between 2hosts

Web browsers do this Example link of rate R

supporting 9 cnctions new app asks for 1 TCP gets

rate R10 new app asks for 11 TCPs

gets R2

Transport Layer 3-102

Delay modeling

Q How long does it take toreceive an object from aWeb server after sendinga request

Ignoring congestion delay isinfluenced by

TCP connection establishment data transmission delay slow start

Notation assumptions Assume one link between

client and server of rate R S MSS (bits) O object size (bits) no retransmissions (no loss

no corruption)Window size First assume fixed

congestion window Wsegments

Then dynamic windowmodeling slow start

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

52

Transport Layer 3-103

Fixed congestion window (1)

First caseWSR gt RTT + SR ACK for

first segment in windowreturns before windowrsquosworth of data sent

delay = 2RTT + OR

Transport Layer 3-104

Fixed congestion window (2)

Second case WSR lt RTT + SR wait

for ACK after sendingwindowrsquos worth of datasent

delay = 2RTT + OR+ (K-1)[SR + RTT - WSR]

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

53

Transport Layer 3-105

TCP Delay Modeling Slow Start (1)

Now suppose window grows according to slow start

Will show that the delay for one object is

Latency = 2RTT +O

R+ P RTT +

S

R

$

amp ( (2P (1)

S

R

where P is the number of times TCP idles at server

P =minQK 1

- where Q is the number of times the server idles if the object were of infinite size

- and K is the number of windows that cover the object

Transport Layer 3-106

TCP Delay Modeling Slow Start (2)

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Examplebull OS = 15 segmentsbull K = 4 windowsbull Q = 2bull P = minK-1Q = 2

Server idles P=2 times

Delay componentsbull 2 RTT for connectionestab and requestbull OR to transmitobjectbull time server idles dueto slow start

Server idles P = minK-1Q times

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

54

Transport Layer 3-107

TCP Delay Modeling (3)

delay =O

R+ 2RTT + idleTimep

p=1

P

=O

R+ 2RTT +

S

R

$ k=1

P

+ RTT amp 2kamp1S

R

( )

=O

R+ 2RTT + P RTT +

S

R

$

( ) amp 2P amp1( )

S

R

S

R+ RTT 2

k1 S

R

$ amp

(

+

= idle time after the kth window

S

R+ RTT = time from when server starts to send segment

until server receives acknowledgement

2k1 S

R= time to transmit the kth window

RTT

initiate TCP

connection

request

objectfirst window

= SR

second window

= 2SR

third window= 4SR

fourth window= 8SR

complete

transmissionobject

delivered

time atclient

time at

server

Transport Layer 3-108

TCP Delay Modeling (4)

K =mink 20S + 21S +L+ 2k1S O

=mink 20 + 21 +L+ 2k1 O S

=mink 2k 1O

S

=mink k log2(O

S+1)

= log2(O

S+1)

$

amp

Calculation of Q number of idles for infinite-size objectis similar (see HW)

Recall K = number of windows that cover object

How do we calculate K

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

55

Transport Layer 3-109

HTTP Modeling Assume Web page consists of

1 base HTML page (of size O bits) M images (each of size O bits)

Non-persistent HTTP M+1 TCP connections in series Response time = (M+1)OR + (M+1)2RTT + sum of idle times

Persistent HTTP 2 RTT to request and receive base HTML file 1 RTT to request and receive M images Response time = (M+1)OR + 3RTT + sum of idle times

Non-persistent HTTP with X parallel connections Suppose MX integer 1 TCP connection for base file MX sets of parallel connections for images Response time = (M+1)OR + (MX + 1)2RTT + sum of idle times

Transport Layer 3-110

HTTP Response time (in seconds)RTT = 100 msec O = 5 Kbytes M=10 and X=5

For low bandwidth connection amp response time dominated bytransmission timePersistent connections only give minor improvement over parallel connections

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo

56

Transport Layer 3-111

HTTP Response time (in seconds)RTT =1 sec O = 5 Kbytes M=10 and X=5

For larger RTT response time dominated by TCP establishment amp slow start delays Persistent connections now give important improvement particularly in high delaybullbandwidth networks

Transport Layer 3-112

Chapter 3 Summary principles behind transport

layer services multiplexing

demultiplexing reliable data transfer flow control congestion control

instantiation andimplementation in theInternet UDP TCP

Next leaving the network

ldquoedgerdquo (applicationtransport layers)

into the networkldquocorerdquo