A Memory Markov Chain Model For VBR Traffic With Strong Positive Correlations.pdf

A Memory Markov Chain Model For VBR Traffic With Strong Positive Correlations O. Rose Institute of Computer Science, University of Wu?rzburg Am Hubland, 97074 Wu?rzburg, Germany Tel.: +49-931-888-5518, Fax.: +49-931-888-4601 Email: rose@informatik.uni-wuerzburg.de WWW: rmatik.uni-wuerzburg.de/～rose We propose a new modeling approach for variable bit rate (VBR) traffic in packet networks based on a Markov chain with memory. The model is simple, comprehensive, and facilitates the modeling of strong positive correlations over a considerable range of lags. The model can easily be adapted to measured traffic sequences. In this paper, this is shown for MPEG video traffic and the output of an IS- 95 vocoder. Due to its simplicity and its close relationship to Special Semi-Markov Processes (SSMPs) the model is valuable to both simulation and analysis of currently considered types of traffic. We present a discrete-time analysis of buffer with Memory Markov Chain input based on a SSMP/D/1-S analysis. 1 Introduction Currently, the increasing amount of traffic types with strong positive correlations or even long-range dependent behavior in combination with larger buffers in telecommunication equipment demand for traffic models and analytical tools capable to cope with this scenario. For instance, if real-time VBR video traffic is considered as being uncorrelated, performance analysis will underestimate losses and delays at network buffers by orders of magnitude [16, 15]. Thus, there is a demand for stochastic processes facilitating the modeling of such correlation charac- teristics. In particular, the modeling of VBR video traffic attracted a lot of interest, e.g. see overviews [7, 17]. Among those modeling approaches, there are Markov chains [13], DARs [10], autoregressive processes [1], TES models [11], selfsimilar processes [8], and combinations of these [5]. The simple 1 modeling approaches, such as histograms or first-order Markov chains, are not capable to provide a good