Kyung Hee University IntServ and DiffServ School of Electronics and Information Kyung Hee University. Choong Seon HONG

Kyung Hee UniversityKyung Hee University

IntServ and DiffServIntServ and DiffServ

School of Electronics and InformationKyung Hee University.Choong Seon HONG<[email protected]>


Quality of Service (QoS)Quality of Service (QoS)

A major driving force in Internet evolutionNot simply defined - means many things to

many peopleHas sense of predictable network behaviourCentral idea is provision of network

resources that an application requires to perform adequately



What is Quality-of-Service?• Quality of service (QoS) is a concept by which applications may indicate and even negotiate their specific service requirements to the network

Why is this an issue?

• The default service in many packet networks is to give all applications the same service, and not consider any service requirements to the network. This is called a best-effort service.



Who needs Quality-of-Service?– Video and audio conferencing bounded delay and

loss rate

– Video and audio streaming bounded packet loss rate

– Time-critical applications (real-time control) bounded delays

– “valuable applications” better service than less valuable applications



How are Quality-of-Service requirements specified?

• QoS parameters are

– Delay

– Delay Variation (Jitter)

– Throughput

– Error Rate



What is the granularity of QoS?

– Per-flow QoS Guarantees are specified and enforced for single

end-to-end data flow

– Aggregate QoSGuarantees are specified and enforced for groups of

flows


Types of QoS guaranteesTypes of QoS guarantees

Deterministic QoS

– Service guarantees are enforced for all traffic For example, deterministic delay guarantees have the form:

Delay of a packet from flow X ≤ D

(D is called a delay bound)

Statistical QoS Allows a certain fraction of traffic to violate the service

guarantees Prob [Delay of a packet from flow X ≤ D ] ≥1 - ε

Where e is a small number (e.g., ε = 10-6)ε


Classification and SchedulingClassification and Scheduling

Routers need to be able to

1. classify arriving packets according to their QoS requirements

Packet Classification

2. isolate traffic flows and provide requested QoS Packet Scheduling


QoS is Generating a Confusing Array of AcronymsQoS is Generating a Confusing Array of Acronyms

Diffserv

QoS

CoS

Intserv RSVP

MPLS

GMPLS


Why Do We Need Such a Revolutionary Change?Why Do We Need Such a Revolutionary Change?

Current ‘best effort’ technology is essentially a quarter of a century old

Two factors driving the development of a new generation of multimedia applications commercialisation of the Internet Increasing availability and decreasing cost of bandwidth

No evidence of ‘free bandwidth’ scenario emerging rejected in RFC1633 (1994) - still true demand always rises to meet supply


QoS is Not NewQoS is Not New

Telephone network has QoS economics and technology based on a single application highly developed engineering but one size fits all

BISDN - an attempt by telephony world to generalise network to encompass diverse applications

ATM technology - first full exploration of QoS on demand concepts


Quality of Service and Resource ManagementQuality of Service and Resource Management

Fundamental resource is output link rateAccess managed via scheduling disciplineBursty input traffic held in buffers

adds delay and jitter overflow causes packet loss

These factors determine QoS at network levelOptimise via buffer management and scheduler

parameter setting


QoS in the InternetQoS in the Internet

Internet Engineering Task Force (IETF) is evolving QoS support mechanisms for the Internet - two approaches The Integrated Services Internet

• QoS for individual microflows

• perhaps too complex for large networks - won’t scale easily

Differentiated Services - more scaleable• lose sight of individual microflows - Behaviour Aggregates


Integrated Services (Intserv)Integrated Services (Intserv)

QoS approached via end to end services best effort - current performance standard controlled load - lightly loaded network performance - ‘s

oft’ delay bound Guaranteed - ‘hard’ bandwidth and delay bounds

Traffic conformance to agreed form expected ‘token bucket’ model - policing if nonconforming

Resources reserved in routers - RSVP more complex set of functions than ATM


RSVP is Dead!RSVP is Dead!

Earlier reports of RSVP’s death were somewhat exaggerated

Nevertheless there is a major problem with Intserv- fatal in the eyes of some

Management of router resources requires each router to maintain per flow ‘state’

Creates ‘state explosion’ in the interior routers of core networks - perhaps confine to edges


Differentiated Services (Diffserv) Differentiated Services (Diffserv)

Driving philosophy of the Internet has been to minimise complexity in the core network - push complexity and intelligence to the edge nodes.

Differentiated Services concept strives to maintain this philosophy while recognising the need to provide some levels of Quality of Service

First widely deployed QoS mechanism


Differentiated ServicesDifferentiated Services


ContentContent

Intserv/RSVPDifferentiated Service ParadigmPer-Hop Behavior & CodepointPremium ServiceAssured Forwarding PHB GroupResource Manager : Bandwidth Broker(BB)Boundary MechanismsDiffserv WG


Internet Integrated Service ModelInternet Integrated Service Model

Guaranteed Quality of Service Motivation

for applications intolerant of late data hard real time requirements

End-to-End behavior an assured level of bandwidth

a delay-bounded service with no queueing loss firm maximum on end-to-end delay not control the minimal or average delay no jitter control


Internet Integrated Service ModelInternet Integrated Service Model In order for a router to invoke Guaranteed Service for a sp

ecific data flow it needs to be informed of the traffic characteristics of the flow, Tspec, along with the reservation characteristics, Rspec Tspec parameters

• p ; peak rate of flow (bytes/second) • b ; bucket depth (bytes) • r ; token bucket rate (byes/second) • m ; minimum policed unit (bytes) • M ; maximum datagram size (bytes)

Rspec parameters • R ; bandwidth, i.e. service rate (bytes/second) • S ; Slack Term (ms)



Controlled - Load Service Motivation

for adaptive real-time applications (today’s internet) work well on unloaded nets but degrade quickly under overload

conditions

--> mimics unloaded nets If the flow is accepted for Controlled-Load Service then the router makes

a commitment to offer the flow a service equivalent to that seen by a best-effort flow on a lightly loaded network


Controlled - Load Service (cont’d) End-to-End behavior

Tightly approximates the behavior visible to applications receiving best-effort service under unloaded conditions

A very high percentage of packets delivered successfully Controlled Load has some fairly simple implementations, in terms of the

queuing systems in routers It is not suited to applications that require very low latency (e.g.

distributed VR systems and so forth).



RSVPRSVP

RSVP Design Principles Receiver-Initiated Reservation

a receiver• choose the level of reservation

• initiate/keep reservation

more flexible and scaleable than source-initiated reservation• heterogeneous receivers

• dynamic membership change

Separating reservation from packet filtering reservation

• amount of resources reserved for an entity

packet filtering• dynamically select packets that can use the resources


Design Principles (cont’d) Maintain “Soft-state”

dynamic status change (membership change) soft-state in switches and maintained by end users state in switches

• path state -- periodic path message from the source• reservation state -- periodic reserv. msg from the receivers

timeout driven deletion• Reservations timeout if not refreshed periodically

adaptability and robustness Protocol overhead

reduce refreshing frequency

merging path/reservation messages

RSVPRSVP


RSVP ATM

Receiver generates reservation Sender generates connection request

Soft state ( refresh / timeout )

Separate from route establishment

QoS can change dynamically

Receiver heterogeneity

Hard state ( explicit delete )

Concurrent with route establishment

QoS is static for life of connection

Uniform QoS to all receivers

Comparison of RSVP and ATM signaling

RSVPRSVP


S1

S2

D1

D2

R1 R2

R3

1. Path_msg

4. forwarding Resv_msg

2. forwarding Path_msg

3-1.Resv_msg

3-2. Resv_msg

RSVP RSVP Message Types


• RSVP designed to work with any protocol - Protocol must provide QoS support - Examples: ATM, IP with Integrated Service

s

• IP integrated services with RSVP over ATM – VC management ( traffic flow-VC)

• Data VC, RSVP signaling VC– QoS translation

• mapping a QoS from the IIS model to a proper ATM QoS

• IIS over POTS• IIS over LAN

Integrated Services over Specific Link Layers(ISSLL WG)

Internet Integrated Service Model


Intserv / RSVP QoS ApproachIntserv / RSVP QoS Approach

Service

- G uaranteed Service - C ontro l Load Service - Best-E ffort Service

Parameters

- Token bucket ru le - Token bucket s ize - Packet rate - M in im al policed unit - M axim um packet s ize

TrafficManagement

- R SVP - Adm ission C ontro l - Q oS R outing - C ontro l o f Traffic Param eters

Scalability problem• Have to maintain forwarding state between receiver and transmitter


Integrated Services ModelIntegrated Services Model

Flow specificationRoutingAdmission controlPolicy controlResource reservationPacket scheduling


RSVP Functional DiagramRSVP Functional Diagram

Application

RSVPD

AdmissionsControl

PacketClassifier

PacketScheduler

PolicyControl

DATA

DATA

RSVPD

PolicyControl

AdmissionsControl

PacketClassifier

PacketScheduler

DATA

RoutingProcess

Host Router


What is a flow?What is a flow?

Equivalent packets by some classification RSVP: Set of packets traversing a network element that

are all covered by the same QoS request

Packet classifier determines which packets belong to which flows IPv6 includes a flow label to ease classification

ISP usage (UUNET) Microflow: TCP or similar bandwidth connection Macroflow: Large aggregates of packets flowing betwe

en two superhubs


Describing and Identifying a FlowDescribing and Identifying a Flow

Flowspec defines traffic parameters Traffic parameters: bandwidth, buffering requirements Uses token bucket specification

Filterspec identifies packets in flow Simplest filter: Source, Dest address/port pair Data filter: classifies packets according to contents


Resource ReservationResource Reservation

Senders advertise using PATH messageReceivers reserve using RESV message

Flowspec + filterspec + policy data Travels upstream in reverse direction of Path message

Merging of reservationsSender/receiver notified of changes


RSVP UDP Reservation (1)RSVP UDP Reservation (1)


RSVP UDP Reservation (2)RSVP UDP Reservation (2)


Client Traffic ShapingClient Traffic Shaping

Issue: Need traffic shaping to meet allocated resources

Source promises that data traffic will conform to a particular shape

Why describe and shape traffic? Network knows what to expect, can manage traffic better Better admission control decisions Network can police flows

Bursty traffic is costly to router, network


Traffic Shaping ExampleTraffic Shaping Example

Flow 1

Flow 2

Data Queue

Data Queue


Traffic ShapersTraffic Shapers

Simple leaky bucket Isosynchronous flow: regular intervals between packets

Token bucket Bursty flow


Simple Leaky BucketSimple Leaky Bucket

Sends data at fixed intervals onto networkBursts bigger than are discardedTraffic never injected faster than Can be used with cells or datagrams

Data

= bucket size= rate data is sent onto network


Token BucketToken Bucket

Sends bursty traffic onto networkBucket filled with tokens at rate Data transmitted when enough tokens exist Allows bursts, but enforces upper bound

Data

= bucket size in tokens= rate tokens are added to bucket

Data Queue


Restrictions on ReservationsRestrictions on Reservations

Admissions Is bandwidth available?

Policy Service guarantees give preferential access to network

bandwidth Permissions Pricing issues

What are the policies of nodes on the path? Policy data represents a scaling and security issue


Resource Reservation ModelResource Reservation Model

Senders advertise using flowspecsRSVP daemons forward advertisements to receive

rs, update available bandwidth, minimum delayReceivers reservations use flowspec, filterspec co

mbination (flow descriptor)Sender/receiver notified of changesReservations are merged in multicast case


Reservation StylesReservation Styles

Wildcard Filter (WF) Shared reservation, select all upstream senders Traffic from upstream senders shares a single pipe Appropriate for audio

Shared Explicit (SE) Shared reservation, explicit sender selection Appropriate for audio

Fixed Filter (FF) District reservations, explicit sender selection Appropriate for video


RSVP FlowspecsRSVP Flowspecs

Peak Data Rate [p]

Minimum Policed Unit [m]

Maximum Policed Unit [M]

Token Bucket Rate [r]

. . .

Token Bucket Size [b]

Sender TSpec, Controlled Load Flowspec

Peak Data Rate [p]

Minimum Policed Unit [m]

Maximum Policed Unit [M]

Token Bucket Rate [r]

. . .

Token Bucket Size [b]

Guaranteed Flowspec

Rate [R]

Slack Term [S]


Packet SchedulingPacket Scheduling

Implemented in hosts/routers to control link allocation

Queuing algorithms Weighted Fair Queuing (WFQ) Class Based Queuing (CBQ)

Queue management Random Early Detection (RED)


Packet SchedulingPacket Scheduling

Fair Queueing Attempts to implement a scheduler that serves all flows

with a backlog at the same rate Emulates a bitwise Round Robin scheduling algorithm Not completely trivial to implement Fair Queuing in a packet netw

ork


Weighted Fair Queuing (WFQ)Weighted Fair Queuing (WFQ)

Traffic placed into queues according to flow specification, flow filter

Fair queuing Implements fairness of bit by bit scheduling on a per

packet basis Gives queues a fair share of total bandwidth

Weighted Queue are not weighted evenly for scheduling

Proven: adequate for Guaranteed Service


Class Based Queuing (CBQ)Class Based Queuing (CBQ)

Combines scheduling and link sharingHierarchical link sharing

Hierarchical queues Enables protocol, organization isolation

Scheduling Does not define a particular scheduling algorithm General scheduler for low latency when no congestion Link-sharing policing scheduler when congested Scheduling per hierarchy


CBQ ExampleCBQ Example

Company B

Real-Time

HTTPFTP

telnet IP DECnet

Company A

Video Audio

60% 40%

30%

20% 10%

20% 10%

LINK

20% 20%


Random Early Detection (RED)Random Early Detection (RED)

Random Early Detection (RED) Queue management algorithm for congestion control Random packet drops as average queue length

increases Can use Explicit Congestion Notification bit instead of

dropping packet Works well for TCP Useful for congested Controlled Load service


Reservation MergingReservation Merging

Receiver#1

Receiver#2

Receiver#3

Reservations mergeas they travel up tree.

R6

R3

R1

R4 R7

(1) 50Kbs

(2) 50Kbs

(3) 50Kbs

(4) 100 Kbs

(5) 100 Kbs

(6) 100 Kbs

(7) 100 Kbs

(8) 60Kbs

(9) 60Kbs


TSpecs, AdSpecs, and RSpecsTSpecs, AdSpecs, and RSpecs

Traffic source sends TSpec (Traffic Specification) Consists of FlowSpec and AdSpec

AdSpec updated to reflect network capabilities Routers update minimum delay and maximum bandwidt

h Termed One Pass With Advertisement (OPSA)

RSpec Receiver uses Controlled Load or Guaranteed FlowSpec

to reserve network resources


Problems with Merging ReservationsProblems with Merging Reservations

Issue: who pays for service, how much?Merging different types of flows

Flow 1: Low delay, low bandwidth Flow 2: High delay, high bandwidth Flow with low delay, high bandwidth satisfies Flows 1

and 2, but it may cost much more than Flow 1 or 2.

Only certain flows can be easily merged given price constraints


Reservation Merging and PriceReservation Merging and Price

Ban

dw

idth

Latency

Reservation 2:High Bandwidth,

High Latency

Reservation 1:Low Bandwidth,

Low Latency

Merged Reservation:High Bandwidth,

Low Latency

Price: Darker = More Costly


RSVP Routing ProblemsRSVP Routing Problems

Routing is separated from admission control If route changes, reservation must be made along

new route New reservation takes time to setup New reservation might fail Old route could still be working fine

Route pinning Always use the route where reservation is in place


Routing Problems (cont’d)Routing Problems (cont’d)

Reservation failure Primary route has inadequate bandwidth although seco

ndary has enough

Telephone system has a crankback feature Allows secondary routes to be considered if reservation

on primary route fails

ATM Routing combined with admission control


Usage and ImplementationUsage and Implementation

RSVP is not widely available Best effort delivery across links with no RSVP services Reservation flag to specify that traffic traveled over a

non-RSVP link

Some links will have guaranteed performance for some traffic, but not all Policy issues at boundaries of networks


Differentiated Service ParadigmDifferentiated Service Paradigm

Complicated operation moves to edge, and stateless in network interior

“ push all the state to the edges, and force all per-conversation work (e.g., shaping, policing) to the edges”

Setting a specific part in an edge node and administrative boundaries DS(differentiated service) field

How to forward according to a specific field of input packet Per-Hop Behavior

According to service rule that is previously promised Traffic Conditioning


Traffic Conditioning

Traffic conditioning mechanisms at the network boundary need to enforce that traffic from a flow adheres to its specification Policing Drop traffic that violates the specification Shaping Buffer traffic at network entrance that violates specification Marking Mark packets with a lower priority or as best effort, if the traffic specification is violated


Traffic Conditioning

The most popular traffic conditioning algorithm is the leaky bucket


Per-Hop Behavior & CodepointPer-Hop Behavior & Codepoint

6-bit DSCP forPer-Hop Behavior

8-bit type ofservice (TOS)

2-bitCU

4-bit version

4-bit header length

16-bit total length (in byte)

Currently Unused

IPv4 Header (first 32bits)

-- DS field in IPv4--


Default PHB

Current best effort forwarding codepoint : 000000

Class Selector PHB

for backward compatibility (IP precedence field) codepoint : xxx000 relative service quality

Per-Hop Behavior & CodepointPer-Hop Behavior & Codepoint


Premium ServicePremium Service

Providing resources according to Peak capacity Pr

ofile Static allocations on peak rate with no statistical sharin

g

Small percentage of the total network capacity allocate

for Premium service

Much higher cost (First class in aircraft)

Commercial applications for Premium service Video broadcasts, voice-over-IP, VPNs, etc.



-- Premium traffic flow from end-host to organization’s ISP --

BorderRouter

BorderRouter

InternalRouter

First-HopRouter

Host

Unmarked packet flow

Packet in premium flows have bit set

Premium packet flow restricted to r rate per sec

Company A

ISP



Forwarding Path Primitives General Classifier

• A transport-level signature matching based on a tuple in the packet header

Bit-pattern Classifier• A simple two-way decision based on whether a particular bit-p

attern in the IP header is set or not

– Ex) ‘P’ bit

Bit setter• Sets the appropriate bits of the IP header to a configured bit-pa

ttern would be the most general



Priority queues (At least) Two levels of simple priority queuing

• The high priority queue for Premium traffic

Shaping token bucket Forward an arriving packet if there is a token present in t

he bucket, otherwise the packet is enqueued until the bucket contains tokens sufficient to send it

• Used in Leaf router

Policing token bucket Never hold arriving packets, but check token availability

• Used in Border router


Assured Forwarding PHB GroupAssured Forwarding PHB Group

N AF classes

M drop precedence level

at this point

4 classes, 3 drop precedence in each class

Example Service : Olympic service


Olympic ServiceOlympic Service

Packets assigned to Gold service class experience lighter load than packets assigned to the silver class

Packets within each class may be further separated by the drop precedence

Drop precedence level by using a dual leaky bucket traffic policer : committed burst, excess burst

Gold Service Class

Silver Service Class

Bronze Service Class


Resource manager Resource manager : Bandwidth Broker (BB): Bandwidth Broker (BB)

A logical entity residing in each administrative domain Managing internal demands & resources according to

the policy database (who can do what when)

setting up & maintaining bilateral agreement with neighbor domains

• bookkeeping how much traffic entering which border router & going out which border router

Today’s BB : network administrators & operators would like to automate over time


Bandwidth Broker (BB)Bandwidth Broker (BB)

Dynamic bandwidth allocation and TCA

management


Choices for implementationChoices for implementation

Adequate provisioning

Manual configuration

not that different from static routing

Using some setup protocols

inter-domain : BB-to-BB

intra-domain : RSVP as a ready candidate


The TCAThe TCAOne per customer - two parts

Constraint TCA protects the provider’s resources

DS field : metering profile : disposition of n/c traffic

quantitative service levels will also include destination address

Fine grain TCA specifies the fine grain traffic conditioning requested by the

customer

mf class. Criteria : mark : shaping profile : disposition of n/c traffic


ConstraintsConstraints

Resources implied by fine grain TCA are constrained by those permitted in the constraint TCA Sum of shaping profiles for each mark must be less than

metering profile for corresponding DS


Inter-domain DS

A Service Level Agreement (SLA) includes a Traffic Conditioning Agreement (TCA)

Simplest way: an administrative issue


Configuring RoutersConfiguring Routers

TCA (Traffic Conditioning Agreement) constraint

fine grain

PHB information

Miscellaneous interface configuration

routing configuration

etc.


Configuring the Constraint TCAConfiguring the Constraint TCA

Specifies agreement between provider and

customer

Relatively static Configuration via SNMP, CLI, COPS, etc.

Dynamic Configuration via COPS or alternate ‘BB’ protocol

Could be triggered by high/low water marks


QoS Policy ControlQoS Policy Control


Configuring the fine Grain TCAConfiguring the fine Grain TCA

Ability to make frequent changes desirable

No need to negotiate with provider

Potentially error prone process

Therefore - use a signaling protocol to do this whenever possible : RSVP

It will still be necessary to provision certain fine grain entries for these, use COPS, SNMP or CLI


Signaled vs. Provisioned Signaled vs. Provisioned TCA EntriesTCA Entries

Quantitative QoS apps use quantitative services TCA entries configured by RSVP signaling or provisio

ned

TCA entries specify egress points

Qualitative QoS apps use qualitative services provisioned only

Should use separate marks (DS-field)


Admission ControlAdmission Control

a/c necessary to prevent over-subscription sum of qualitative entry shaping profiles leq than meter

ing profile for qualitative DS fields• enforced at provisioning time(SNMP, CLI or COPS)

sum of quantitative entry shaping profiles leq than metering profile for quantitative DS fields

• enforced by rejection or acceptance of RSVP signaling messages

• if quantitative resources are provisioned, a/c must be enforced both at provisioning and signaling times

• difficult to manage


Diffserv WG statusDiffserv WG status

Closed

RFC 2474, Definition of the Differentiated

Services Field (DS Field) in the IPv4 and IPv6

Headers

RFC 2475, An Architecture for Differentiated

Service

---


Diffserv / MPLSDiffserv / MPLS

Diffserv BA (Behavior Aggregate)

• BAS (BA Selector) ; forwarding queue behavior

• BM (Behavior Modifier) ; Dropping behavior

PHP

MPLS Single LSP

• Same BAS

• BM Carried as part of label encapsulation header


Diffserv / RSVPDiffserv / RSVP

Diffserv NetworkPEP

PEP PEP

PEP : Policy Enforcement PointPEP : Policy Enforcement Point

PDP : Policy Decision PointPDP : Policy Decision Point

PEP

PEP PEP

PDP

Diffserv / PHP / CodepointDiffserv / PHP / CodepointRSVPRSVP RSVPRSVP

Policy ServerPolicy Server

Documents

Kyung Hee University IntServ and DiffServ School of Electronics and Information Kyung Hee University. Choong Seon HONG