简介:InInternetenvironment,trafficflowtoalinkistypicallymodeledbysuperpositionofON/OFFbasedsources.DuringeachON-periodforaparticularsource,packetsarriveaccordingtoaPoissonprocessandpacketsizes(henceservicetimes)canbegenerallydistributed.Inthispaper,weestablishheavytrafficlimittheoremstoprovidesuitableapproximationsforthesystemunderfirst-infirst-out(FIFO)andwork-conservingservicediscipline,whichstatethat,whenthelengthsofbothON-andOFF-periodsarelightlytailed,thesequencesofthescaledqueuelengthandworkloadprocessesconvergeweaklytoshort-rangedependentreflectingGaussianprocesses,andwhenthelengthsofON-and/orOFF-periodsareheavilytailedwithinfinitevariance,thesequencesconvergeweaklytoeitherreflectingfractionalBrownianmotions(FBMs)orcertaintypeoflongrangedependentreflectingGaussianprocessesdependingonthechoiceofscalingasthenumberofsuperposedsourcestendstoinfinity.Moreover,thesequencesexhibitastatespacecollapse-likepropertywhenthenumberofsourcesislargeenough,whichisakindofextensionofthewell-knownLittle’slawforM/M/1queueingsystem.Theorytojustifytheapproximationsisbasedonappropriateheavytrafficconditionswhichessentiallymeanthattheserviceratecloselyapproachesthearrivalratewhenthenumberofinputsourcestendstoinfinity.