资源描述:
《Time Series Models for Internet Traffic》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、TimeSeriesModelsforInternetTraffic*SabyasachiBasuAmarnathMukherjeeDepartmentofStatisticalScienceSteveKlivanskySouthernMethodistUniversityCollegeofComputingDallas,TX75275GeorgiaInstitUteofTechnologyAtlanta,GA30332AbstractoflengthT.Thekeyobservationonwhichthem
2、od-elsarebasedisthatappropriatedifferencedsequencesDatatraficsequencesfromtwocampusFDDIrings,{Wt=yt-yt-s},(s>_l),arestationaryprocessesanEthernet,twoentry/exitpointsoftheNSFNET,withshortmemory,andcanassuchbemodeledasandsub-sequencesbelongingtopopularTCPportn
3、um-Auto-RegressiveMoving-Average(ARMA)processesbersononeoftheFDDIringsindicatethatappro-withorders(p,q),wherepandqaresmallintegers.priatelydifferencedtime-seriesgeneratedfromtheseFormostofthedatasetswehavestudied,swasequaltracescanbemodeledasAuto-Regressive-
4、Moving-tooneortwo.Fortwodatasets,itsvaluewaseightforAverage(ARMA)processes.Thevariatesoftheonetimeinterval(T),indicatinganon-randomperiod-ARMAfilterare,however,non-Gaussian.icityoflength8,butoneatasmaller(andalsolarger)Asequenceofstepsleadingthrough(i)parame
5、tertimeinterval.Alistofdatasetsstudiedarelistedinestimation,(ii)generatingthedistributionofthevari-Table1.(FordetailsofTCPtrafficandtheirmodels,ates,(iii)forecastingtailpercentiles,and(iv)syntheticsee[l].)Spectraldensitiesoftwodatasets(giveningenerationofnon
6、-negativeintegersequencesispre-[l]),however,showquiteafewnon-randomperiodici-sented.Thedataindicatesthatparameterestimatesties,andareanexceptiontotheaboverule.driftslowlywithtimeandmayneedtobere-computedBywayofexample,considertheautocorrelationperiodicallyfo
7、raccurateforecasts.Theforecastingal-function(ACF)oftwoofthedatasets.First,inFig-gorithmhaspotentialapplicationindynamicresourceurel(a),weseetheACFofatraceovertheSanDiegoallocation.ThesynthetictrafficgenerationalgorithmexternalinterfacetotheNSFNET(ENSS),studi
8、edinmaybeusedinsimulationstudiesofresourcemanage-[4],withtimeinterval,T,setto0.1seconds.InFigurementalgorithms.l(b),theACFof{yt-Yt-l}isseentohaveonlyafewsignificantauto-correlationsleft;thesewer