AQUILA (IST-1999-10077) - TU Dresden fileAdaptive Resource Control for QoS Using an IP-based Layered...

Adaptive Resource Control for QoSAdaptive Resource Control for QoSUsing an IPUsing an IP--based Layered Architecturebased Layered Architecture

AQUILA

Bert F. KochBert F. Koch

StefanoStefano SalsanoSalsano

http://wwwhttp://www--stst.inf..inf.tutu--dresdendresden.de/.de/aquilaaquila//

2001 Tyrrenian International Workshop on Digital Communications,IIWDC 2001, Taormina, Italy, September 2001

Evolutionary Trends of the Internet

IP QoS at work: Definition and IP QoS at work: Definition and Implementation of the AQUILA ArchitectureImplementation of the AQUILA Architecture

(IST-1999-10077)

(c) 2001 AQUILA consortium. IWDC 2001, 17-20.09.20012

AQUILA

Outline

nn Project IntroductionProject Introduction

nn AQUILA QoS ArchitectureAQUILA QoS Architecture

nn Traffic Engineering AspectsTraffic Engineering Aspects

nn IP QoS at work: the trial IP QoS at work: the trial

nn (SLS (SLS -- Service Level Specification)Service Level Specification)

nn Ongoing activitiesOngoing activities


AQUILA

Outline




nn Further Project ActivitiesFurther Project Activities




AQUILA

ConsortiumSAGSAG Siemens (CoSiemens (Co--ordinatorordinator), Germany), Germany

BAGBAG Bertelsmann Bertelsmann mediaSystemsmediaSystems, Germany, GermanyDTADTA TT--Nova Deutsche Telekom, GermanyNova Deutsche Telekom, GermanyTAATAA Telekom Telekom AustriaAustria, , AustriaAustriaELIELI Elisa Elisa CommunicationsCommunications, , Finland Finland TPSTPS PolishPolish Telecom, PolandTelecom, Poland

NTUNTU National National TechnicalTechnical University of Athens, University of Athens, GreeceGreeceWUTWUT WarsawWarsaw University of Technology, PolandUniversity of Technology, PolandCORCOR CoRiTelCoRiTel, , Italy Italy TUDTUD Dresden University of Technology, GermanyDresden University of Technology, GermanySPUSPU Salzburg Research, Salzburg Research, AustriaAustria

QSYQSY QQ--Systems, Systems, GreeceGreece

I&CI&Cmanufacturermanufacturer

Internet ServiceInternet ServiceProvidersProviders

andandNetworkNetwork OperatorsOperators

UniversitiesUniversitiesandand

ResearchResearchInstitutesInstitutes

Web Web applicationapplicationproviderprovider


AQUILA

Main Innovations

ResourceControl Layer

DiffServDomainISP A

ISP Internet Service Provider

RCL Resource Control Layer

DiffServ DomainISP B

ED

Inter-Domain QoS

RCL

Scalable and flexible Admission Control and Resource Management

End-user Application Toolkit to request QoS

Host

Host

EdgeDevice

EdgeDevice

CoreRouter

CoreRouter

CoreRouter

EdgeDevice

Web-Server

Distributed QoS Measurement

⇒ end-to-end Quality of Service

Distributed QoS MeasurementTraffic Control

Algorithms


AQUILA

Design Goals

nn Scalable ArchitectureScalable Architecture• Distributed building blocks• Autonomous operation of elements

nn Based onBased on DiffServDiffServ Network ElementsNetwork Elements• Use of existing, commercial routers• Enable migration path from current networks

nn Resilience to failuresResilience to failures• Failure of an element should only degrade (if at all), not disable the

operation of other elements


AQUILA

Outline




nn IP QoS at work: the trial IP QoS at work: the trial

nn (SLS (SLS -- Service Level Specification) Service Level Specification)

nn Future PlansFuture Plans


AQUILA

Resource Control Layer (RCL)

nn Tasks of the Resource Control LayerTasks of the Resource Control Layer• Admission Control to limit the amount of prioritised traffic• Resource Management• QoS Interface

nn Design GoalsDesign Goals• Simpler than ATM / RSVP: no explicit reservation along the

data path• Scalable approach for Admission Control (distributed

Admission Control, separation of Admission Control and Resource Management)


AQUILA

Scalable Architecture for RCL

ISP Domain

Edge Router

Core Router

Core Router

Core Router

AccessNetwork

Edge Router

Res

ou

rce

Co

ntr

ol L

ayer

AccessNetwork

Admission ControlAdmission Control

QoS Req

uest S

ettin

gsAdmission

Control Agent

Set

tings

Admission Control Agent

QoS Request

QoS Request

setti

ngs

(rou

ting.

..)

Used resources

resources

Resource Control Agent

Resource ControlResource ControlConsideration of Network LoadConsideration of Network Load Monitoring

ProbingResultsRCA

ACA ACAEAT EAT

RCA: Resource Control AgentACA: Admission Control Agent

EAT: End user Application Toolkit RCL: Resource Control Layer

CORBA based communications


AQUILA

Resource Control Agent (RCA)

nn Manages resourcesManages resources

nn Checks availability of requested resourcesChecks availability of requested resources

nn (Re(Re--)distributes resources as needed)distributes resources as needed


AQUILA

Admission Control Agent (ACA)

nn Authenticates userAuthenticates user

nn Authorises and checks requestAuthorises and checks request

nn Locates ingress and/or egress edge routerLocates ingress and/or egress edge router

nn Requests resources from the resource control agentRequests resources from the resource control agent

nn Admits / rejects new flowsAdmits / rejects new flows

nn Installs policies in ingress routerInstalls policies in ingress router


AQUILA

End-user Application Toolkit (EAT)

nn Middleware between QoS Network and ApplicationMiddleware between QoS Network and Application• Front end for access network• QoS portal for application (legacy and QoS aware)• Alternative, flexible approach for evaluating QoS reservations

nn EAT is requester of QoS reservationEAT is requester of QoS reservation• The requester may be the sender, receiver or a third party• The requester initiates the reservation• The requester is charged for the service


AQUILA

Objectives of EAT

nn Enable Access to QoS to non Enable Access to QoS to non QoS aware QoS aware Legacy ApplicationsLegacy Applications

nn Support Support QoS aware ApplicationsQoS aware Applications(RSVP,(RSVP, DiffServDiffServ) / Support ) / Support various QoS Request Methodsvarious QoS Request Methods

nn Provide a Methodology and a Provide a Methodology and a Programming Interface to Programming Interface to support the Construction of support the Construction of new new QoS aware ApplicationsQoS aware Applications

nn Provide an Provide an EndEnd--user friendly user friendly QoS AccessQoS Access

Pro

xies

AP

I Network

Control plane

Data plane

End-user Application Toolkit

QoS

ACA


AQUILA

Outline




nn IP QoS at work: the trialIP QoS at work: the trial




AQUILA

Network Services and Traffic Classes

Network ServicesNetwork Services

CustomerCustomer

Network operatorNetwork operator

Traffic ClassesTraffic Classes

The network operator offersthe NS to the customer

Network services are mappedinto traffic classes

DSCP, scheduling and queuingalgorithms (e.g. WFQ, RED), router

configuration, admission control rules


AQUILA

Network Services

nn CharacterisationCharacterisation• Delivery of services to the customer• Defined by the network operator• Provides a specific QoS, expressed by statistical or deterministic

statements about delay, loss, ...

nn ImplementationImplementation• Network service are implemented by traffic classes

nn UsageUsage• The End-user Application Toolkit maps application demands to

network services

nn Example: Premium CBR for IP Telephony and VoiceExample: Premium CBR for IP Telephony and Voice TrunkingTrunking

⇒ Goal: only a few Network Services to allow clear service Differentiation (wrt QoS objectives)


AQUILA

Network Services

nn PCBR Premium Constant Bit RatePCBR Premium Constant Bit Rate

• VoIP,Voice Trunking, VLL

nn PVBR Premium Variable Bit RatePVBR Premium Variable Bit Rate

• Video (real time)

nn PMM Premium MultimediaPMM Premium Multimedia• Video (download), file sharing, web

nn PMC Premium Mission CriticalPMC Premium Mission Critical• Transactional application, interactive games

nn Best EffortBest Effort


AQUILA

Traffic Classes

nn Five Traffic Classes have been specifiedFive Traffic Classes have been specified

nn … as well as the related Traffic Control Mechanisms in the … as well as the related Traffic Control Mechanisms in the RoutersRouters

Networkservice

Premium CBR Premium VBR PremiumMultiMedia

PremiumMission Critical

Standard

Traffic class TCL 1 TCL 2 TCL 3 TCL 4 TCL STD

PQPacketarrives

TCL 3

TCL 2

TCL 4

TCL 1

Classifier

Packetdeparts

WFQ

TCL STD

High priority

Low priority


AQUILA

Overview of Traffic Handling Approach

PROVISIONING

Configuration ofTRAFFIC

CONTROL

Limits forADMISSIONCONTROL

First trial: feed-forward operations,there are no “control loops”


AQUILA

Admission Control

nn Declaration Based Admission Control (First Trial)Declaration Based Admission Control (First Trial)Traffic Characterisation based on token bucket parametersTraffic Characterisation based on token bucket parameters

nn Admission Control AlgorithmsAdmission Control Algorithms

take into account the traffic parameters specified in the take into account the traffic parameters specified in the reservation request and compare with the provisioned admission reservation request and compare with the provisioned admission control rate limits and the already allocated resources.control rate limits and the already allocated resources.

nn Specific Admission Control AlgorithmsSpecific Admission Control Algorithms

have been defined for the different traffic classes and differenhave been defined for the different traffic classes and different t (high speed / low speed) access links.(high speed / low speed) access links.


AQUILA

Provisioning

nn Initial ProvisioningInitial Provisioning

Constraints on Traffic Classusage per Link

Expected traffic matrix

Topology and routing

Provisionedrates per

link

nn Building Resource PoolsBuilding Resource Pools• Resource pools are built when it is useful to dynamically share a

bottleneck link among a set of access links

ACrate limits

Initial provisioningalgorithm


AQUILA

Resource Pools

nn Resource LimitsResource Limits• Limit amount of QoS

traffic from each edge router

nn Group neighboured Group neighboured RoutersRouters• Limit amount of QoS

traffic from each group

nn Dynamic DistributionDynamic Distribution• Dynamically shift

resources within group

nn Hierarchical StructureHierarchical Structure• “Groups of groups”

Domainsub-area

subordinatedsub-area


AQUILA

Dynamic aspects

nn The resource reservations are dynamicThe resource reservations are dynamic

nn The handling of network resource is dynamic, but The handling of network resource is dynamic, but

distributed and scalabledistributed and scalable


AQUILA

E2E E2ESynthetic End

Synthetic End--

toto--End Load

End Load

Measurements

Host Host

CoreRouter

measurementdatabase

test description

ProbeProbe

ProbingProbing

EdgeDevice

EdgeDevice Core

RouterCore

Router

MonitoringMonitoring

ResultsResults

Reser-vation

EAT

Resource ControlTraffic Control

Admission Control


AQUILA

Outline




nn IP QoS at work: the trialIP QoS at work: the trial




AQUILA

Trials at Three Different Sites

nn Warsaw (Polish Telecom)Warsaw (Polish Telecom)• Reference Site• Special Focus on Streaming Media / Video on Demand

nn Helsinki (Elisa Communications)Helsinki (Elisa Communications)• Site with various Access Technologies (ADSL, 10 Mbps Ethernet, WLAN) • Special Focus on realistic customer/end-user environments and different

environments and usage (home, office, public access zone)nn Vienna (Vienna (TelekomTelekom Austria)Austria)

• Site with homogenous Layer 2 (Ethernet)

• Special Focus on low bandwidth real-time applications, VoIP, Multi-user network games


AQUILA

Test-bed in Warsaw

Internet

aq_3640_4

aq_7507_1

aq_7507_3

aq_7507_2

aq_3640_2

aq_3640_3aq_1605_2

PC5GPS

PC6

PC8PC7

PC1GPS

PC2

SUN1 SUN2 SUN3

RCAACAEAT

MeasurementServer

PC3GPS

PC4GPS

155 Mbps

155 Mbps

10 Mbps

10 Mbps

2 Mbps

2 Mbps

2 Mbps

AntennaGPS

aq_3640_1


AQUILA

Outline








AQUILA

nn In the AQUILA architecture a “Reservation Request” is sent to In the AQUILA architecture a “Reservation Request” is sent to the Admission Control Agent (ACA) by the “Endthe Admission Control Agent (ACA) by the “End--user user Application Toolkit” (EAT)Application Toolkit” (EAT)

AQUILA Approach to SLS

H – Host

QMTool

Core DiffServ Network

Application

EAToolkit

HAccessNetwork ED

BR

ED

CR

CR

CR

CR

ISP

AccessNetwork H

ED – Edge DeviceBR – Border RouterCR – CoreRouter

Reservation Request RCA

ACAEAT

nn A Reservation Request contains the Service Level SpecificationA Reservation Request contains the Service Level Specification

ACA

ACA


AQUILA

Semantic Content of SLS

see draft-salsano-aquila-sls-00.txt

nn SLS typeSLS type

nn ScopeScope

nn Flow IdentificationFlow Identification

nn Traffic description and Traffic description and conformance testconformance test

nn Performance GuaranteesPerformance Guarantees

nn Service scheduleService schedule


AQUILA

Outline








AQUILA

Ongoing activities

nn InterInter--Domain scenariosDomain scenarios

• Extension of “Border Gateway Reservation Protocol” BGRP proposal• SLA/SLS between ISPs

nn MeasurementMeasurement

• From Declaration-Based Admission control to Mixed Declaration and Measurement Based Admission Control

• Resource control loops at the provisioning level (feedback from measurement to resource control)


AQUILA

Ongoing activities

nn Support of ApplicationsSupport of Applications• RSVP as QoS signalling protocol• Application Programming interface (API)

nn ManagementManagement• Administration of network services• Control of resource distribution parameters

nn Architectural extensionsArchitectural extensions• Impact of MPLS

– VPN– MPLS-DiffServ interworking


AQUILA

PROVISIONING

Control Loops

Measurements of:- Traffic

- QoS parameters

A

B

CLimits for

ADMISSIONCONTROL

Settings forRESOURCE

POOLS

Configuration ofTRAFFIC

CONTROL

Sharing policies

Expected traffic

Topology and routing


AQUILA

Conclusions

nn IP QoS is usefulIP QoS is useful

• Why ? In order to differentiate between applications (e.g. voice/data)In order to provide different level of services

• Where ? Only on the bottleneck links: at network access

nn ScalableScalable dynamidynamic IP QoS is feasible (single domain)c IP QoS is feasible (single domain)

• We are working on the inter-domain aspects

(IST-1999-10077)

Adaptive Resource Control for QoSAdaptive Resource Control for QoSUsing an IPUsing an IP--based Layered Architecturebased Layered Architecture

AQUILA

http://wwwhttp://www--stst.inf..inf.tutu--dresdendresden.de/.de/aquilaaquila//

Thank you forThank you foryour attention !your attention !

AQUILA (IST-1999-10077) - TU Dresden fileAdaptive Resource Control for QoS Using an IP-based Layered...

Documents

Transcript of AQUILA (IST-1999-10077) - TU Dresden fileAdaptive Resource Control for QoS Using an IP-based Layered...