ON THE IMPACT OF LINKNODE FAILURES AND NETWORK APP

PERIODICAPOLYTECHNICASER.EL.ENG.VOL.46,NO.1–2,PP.99–122(2002)ONTHEIMPACTOFLINK/NODEFAILURESANDNETWORKAPPLICATIONSONTHELOADANDCALLPROCESSINGTIMESINATMNETWORKSSándorSZÉKELY1,GáborSZ˝UCS1,CsabaSIMON,IstvánMOLDOVÁNandSándorMOLNÁRHighSpeedNetworksLaboratoryDepartmentofTelecommunicationsandTelematicsBudapestUniversityofTechnologyandEconomics1117Budapest,Magyartudósokkrt.2,I.B.210,HungaryTel:[phone],fax:[phone]e-mail:{szekely,szucs,simon,moldovan,molnar}@ttt-atm.ttt.bme.huReceived:December10,2002AbstractThebeginningofthe1990sbroughtnewtechnologiesinthetelecommunicationnetworks.Asyn-chronousTransferMode(ATM)hasbeenchosenasthetransmissiontechnologyofbroadbandnet-works.ThestandardisationofATMispracticallyfinishedtoday,theimplementationsarereadyandtheATMequipmentislargelyintroducedbyusersandserviceproviders.Attheendofthe1990stheInternetProtocol(IP)becamethefastestgrowingnetworklayerprotocolthatisapplicableoveranydatalinklayer.Theconvergenceofthesetwotechnologiesbecamerealityatthebeginningoftheyear2000,butacoupleofmonthslaterthetelecommunicationmarketenteredthedeepestreces-sioneverseen.Inthecurrentmarketplacetheserviceprovidersmustimprovenetworkmanagementtoreduceoperationcosts.ATMnetworkserviceprovidersofferedmainlypermanentvirtualcircuitconnectionstocustomersinthelastyears,butrecentlythereisanincreasinginteresttoofferswitchedvirtualcircuit(SVC)connectionstoendusers.TheSVCisbasedontheuseofsignallingprotocols.OurpaperfocusesontheperformanceofcallprocessinginATMnetworks.BasedonaseriesofmeasurementsonfourtypesofATMswitcheswehaveestablishedsomegeneralisedconclusions,whichareswitchinvariantfeaturesoftheATMsignallingflows.Theseresultshaveawidegeneral-ityamongATMswitchesandtheycanbeusedtomodelanddesignlargeATMsignallingnetworks.Ournewmodelisvalidatedagainstthemeasurementresultsonthe7-nodebackboneoftheTEN-155Pan-EuropeanATMnetwork.Thesecondpartofthepapershowsthatacascadednetworkisagoodestimatorforthesignallingperformanceofanarbitrarynetwork.Inaddition,itisdemonstratedthatthecalldensityofthenetworkisanimportantnetworkparameter,whichiscloselyrelatedtothemaximumnetworklevelcallarrivalrate.Theimpactofthelink/nodefailuresandnetworkapplica-tionsonthecallestablishmenttimesareinvestigatedinthisarticleona35-nodesamplenetwork.Thecasestudiesareextendedfromhomogeneousnetworkstohybridnetworks,wheremanytypesofswitchesarepresent.Keywords:ATMsignalling,performancemeasurements,modelling,networkdiameter,averagecallpath,calldensityofthenetwork,maximumnetworklevelcallarrivalrate,linkandnodefailures.1TheauthorsarecurrentlywithSiemensAG,ICN,Munich,Germany100S.SZÉKELYetal.1.IntroductionOneoftheadvantagesofATMistheabilitytosetupandteardownvirtualcon-nectionsdynamicallybetweensourceanddestination.Therearemanyfactorsthatmayinfluencethenumberofswitchedconnectionsanetworkcanacceptandtherateatwhichtheycanbeaccepted.Bothoftheseperformancecriteriaareinflu-encedbytheUser-Network-Interface(UNI)signallingperformance[1].AgoodsurveyofsignallingprotocolsforATMnetworkscanbefoundin[24],[18]and[3].WiththeincreasedcapabilityandcomplexityofATMnetworks,callprocess-ingandespeciallysignallingproceduresbecomemoreandmorecomplexwhichcausesaperformancedegradationofthesignallingnetworks.Moreover,thesizeofthesignallingmessagesislargerandlarger.ThereforetheexistingmodelsforSS7networks,e.g.[2],donotrepresentcorrectlythebehaviourofthesignallingprocessinginbroadbandnetworks.Thereisaneedfornewsignallingmodels.ThebasicperformancemetricsforATMsignallingaredescribedin[1],whileperfor-mancebenchmarkingofATMsignallingsoftwareispresentedin[11]and[16],showingthefirstresultsobtainedonswitchesfromfourvendorsinanumberofhardware,softwareandnetworkconfigurations.[16]hassomeearlyresultsonmultiplehostconnectionsaswell,butthereismoreworktobedone.AlthoughtheimportanceofthesignallingperformanceofATMnetworkshasbeenrecognisedasapotentialbottleneckin[6],veryfewpapersaddressedthecongestionsituationinswitchesduetosignallingmessageflows.Furthermore,theevaluationin[12]isbasedonestimationsandexternalinputsandexpressesthenecessitytohaverealmeasurementsoftheequipmentthatisgoingtobeinstalledforBroadband-Video-on-Demand(B-VoD).BasedontheseandmanyotherstudieswehaveconcludedthattheexistingmeasurementresultsofcurrentbroadbandsignallingnetworksarefarnotenoughtocoverallaspectsoftheATMsignalling.Motivatedbythisneedweaddressedthisresearchareaandthereforethefirstpartofourpaperfocusesonrealmeasure-mentsofsuchsignallingnetworks.Section2describesourobjectives,themethodologywehaveusedduringourevaluationandestablishesourdefinitionsusedinthisarticle.Section3focusesontheperformanceofthecallprocessingbymeasurementsofisolatedswitches.TheaimistoidentifythemaincomponentsofcallprocessinginATMnetworksandtodescribequantitativelytheeffectofdifferentcallprofiles.Section4dealswiththeconstructionandperformanceanalysisofanewgenericflowmodelbasedonthesemeasurementresultsfromSection3.Theanalysisisextendedtonetworklevelinsection5,differentnetworktopologiesareinvestigated:10-nodecascaded,4-nodefullymeshed,7-node,30-nodeand35-nodearbitra