An Empirical Approach For Multilayer Traffic Modeling And Multimedia Traffic Modeling At Different Time Scales

Abstract

We describe empirical approaches for multilayer traffic modeling — i.e., models that span several protocol layers — and for modeling multimedia traffic at various time scales.

Multilayer traffic modeling is challenging, as one must deal with disparate traffic sources; control loops; the effects of network elements; cross-layer protocols; asymmetries in bandwidth, session lengths, and application behaviors; and an enormous number of potential confounding effects among the various factors.

We summarize experiments that combine an analytical transport layer model (layer 4) with layer 1/2/3 components to investigate whether analytical multilayer traffic models might provide credible outcomes in (near) real time. Preliminary results suggest that such models can provide reasonable, steady-state, first-order approximations of behaviors that span several protocol layers.

Multimedia traffic modeling is also challenging, as many types of multimedia traffic have characteristic statistical signatures induced by their encoders. Traffic analysts have proposed a number of feasible models for multimedia traffic, but it is not clear which is best.

We summarize experiments using multiplicative SARIMA(s,p,d,q) models of MPEG-4 multimedia traces at various time scales. Preliminary results suggest that the seasonal effect induced by MPEG’s ‘group of pictures’ encoding is the dominant factor at time scales up to a few tens of seconds, while scene length predominates at longer time scales.