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