Can self-similar traffic be modeled by Markovian processes?

Abstract

In this paper, we compare high time resolution local area network (LAN) traffic with three different traffic models: Poisson, ON-OFF and 5-state Markov process. Due to the measured data's extreme variability on time scales ranging from milliseconds to days, it is difficult to find a model for it, especially a Markovian one. Recent studies show that conventional models do not capture the characteristics of the observed traffic. Fractal-based models have already been built to characterize such a traffic but they are not easily tractable tractability of them is not great. Through a new method which integrates different time scales in the model, we have tried to find a quite simple Markovian process having the same behavior as the measured traffic on the LAN. We show in particular that a simple 5-state Markov process integrating different time scales can reasonably model the behavior of measured traffic up to a certain time interval.KeywordsLAN Trafficself-similarMarkovian modelslong range dependences