Performance Analysis of Reactive Congestion Contro

PerformanceAnalysisofReactiveCongestionControlBaseduponQueueLengthThresholdValuesM.KatoaY.OiebM.MuratacH.MiyaharacaDepartmentofElectronicsandInformationEngineering,AriakeNationalCollegeofTechnology,Omuta,836JapanbInformationTechnologyCenter,NaraInstituteofTechnology,Ikoma,630-01JapancDepartmentofInformationandComputerSciences,FacultyofEngineeringScience,OsakaUniversity,Toyonaka,560JapanAbstractReactivecongestioncontrolbasedupontwothresholdsisanalyzed.Underthecongestioncontroltreatedhere,amultiplexerortransitnodeinthenetworkrecog-nizescongestionwhenitsqueuelengthreachessomepre-determinedthresholdvalue,denotedbyH.Thecongestednodethenexplicitlyinformsthesourcessharingthenodeofitscongestioninsomewaytomarkaforwardcongestionnotificationbitintheheaderofcellsifprovided,ortodirectlysendcontrolcellstothosesources.Thesourcesthenstopthecelltransmissionaccordingtothefeedbackinformationreceived.Thereliefofcongestionisrecognizedbythenodewhenitsqueuelengthdecreasestosomelow-threshold,L,andinformsthesourcesthatthesourcescanagainstartthecelltransmission.Inthispaper,weinvestigatethissortofcongestioncontrolbyformingaqueueingnetwork,inwhichmultiplesourcesandasinglenodeareexplicitlymodeled.Throughananalysis,wediscusstheimpactofthosetwothresholdstoperformanceofthecongestioncontrol.Keywords:ATM,ATMLAN,congestioncontrol,reactivecongestioncontrol,explicitcongestionnotification,performanceanalysis1IntroductionInbroadbandintegratedservicesdigitalnetworks(B-ISDN),asynchronoustransfermode(ATM)isemployedtoofferawidevarietyofservices(forex-ample,data,voice,videoetc).InATMnetworks,congestioncontrolisveryimportantforachievinghighutilizationofthenetworkandguaranteeingtheQOS(QualityOfService)requiredbyvariousservices.CongestioncontrolinATMnetworksisdividedintotwocategories:preventivecontrolandreactivePreprintsubmittedtoElsevierScience9June2000control.InwideareaATMnetworks,preventivecontrolhasbeenemployedbecauseitisrecognizedasamethodtoefficientlytreatcongestion[3].Underthepreventivecongestioncontrol,usersarerequiredtodeclarethecharacter-isticsoftheirtrafficinadvanceforconnectionsetup,andthenthenetworkdetermineswhetheritpermitstherequestedcallornot.Afterthecallestab-lishmentrequestisadmitted,theuserisnotallowedtoviolateitsdeclaredtrafficparametersduringtransmission.Bythis“trafficcontract”mechanism,QOScanbeattainedforeachuser.Thus,itshouldbesuitabletorealtimetrafficsuchasvoiceorvideoinwhichitstrafficcharacteristicscanbeknowninadvance.Ontheotherhand,whendatacommunicationsisconsidered,itisnotverylikelythatusersknowthecharacteristicsoftheirtrafficinadvance.Amoreimportantproblemisthatinpreventivecongestioncontrol,theusersnevertransmitcellsexceedingthetrafficcontractevenwhenthenetworkisnotcon-gested.Therefore,itisapparentthathighspeeddatacommunicationisnotachieved.Thus,inATMlocalareanetworks(LANs),anotherkindofcon-gestioncontrolhasbeenstudied,i.e.,reactivecongestioncontrol[5].Underthereactivecongestioncontrol,atransitnodeinthenetworkrecognizescon-gestionoccurringbydetectingthatitsqueuelengthreachessomehigh-levelthresholdH.Thecongestednodetheninformsthesources,whichsharethelinkofthatnode,inoneofthefollowingtwoways.Thenodemarksaforwardcongestionnotificationbitintheheaderofcellsifprovided,andthosecellsarriveattheirdestinations.Eachdestinationtheninformsitscorrespondingsourceofthecongestionbymarkingabackwardcongestionnotificationbitofthebackwardcells.ThistypeofcontroliscalledFECN(ForwardExplicitCongestionNotification)basedcongestioncontrol.Analternativeapproachisthatthecongestednodegeneratesanddirectlysendscontrolcellstothesources,whichiscalledtheBECN(BackwardECN)basedcongestioncontrol.Thereliefofcongestionisnotifiedwhenthequeuelengthdecreasestosomelow-levelthreshold,L.Notethatthesetwowaysaredifferentonlyinthedelayexperiencedbyfeedbackinformationinthecurrentstudy.In[1,2,4,7],reactivecongestioncontrolshavebeenstudied.Kawaharaetal.haveanalyzedin[7]theimpactofburstytraffictotheperformanceofthereactivecongestioncontrolwhereasthelow-levelthresholdvalueisfixedatH−1.Ontheotherhand,Wangetal.[1]haveanalyzedthereactivecongestioncontrolbasedupontwothresholds.However,cellsareassumedtobealwaystransmittedaccordingtoaPoissondistributioninavailableslotsfromeachsource.Referto[7]forotherrelatedstudies.Inthispaper,wedevelopaqueueingnetworkmodeltoanalyzethereactivecongestioncontrolbasedonECNschemesdescribedabove.Ourmodelisde-velopedinawaytoextendthemodelof[1]totreatmorerealisticcaseinwhicheachsourceisequippedwithbufferstostorecellswhenthecelltransmission2isprohibited.Furthermore,themodelenablesustodiscusstheeffectoftwothresholdsontheperformanceandalsotoevaluatetheamountofcontrolcellsrequiredintheBECNscheme.Theremainderofthispaperisorganizedasfollows.InSection2,wedescribetheanalyticalmodelandthereactivecongestioncontrolschemeconsideredhere.Section3presentsouranalysisofthereactivecongestioncontrolandthederivationofthecelllossprobabilityandtheaveragecyclelength.InSection4,wediscusstheperformanceofthecongestioncontrolthroughnumericalresults.Finally,Section5summarizesthepaper.2ReactiveCongestionControlSche