Study On Self-Similar Traffic And Its Influence On Network Performance

A number of recent measurements and studies of actual traffic from working networks (LAN and WAN) demonstrated that real traffic has statistical self-similarity and, therefore, is long range-dependent. These studies prove that traditional models neglect this importance characteristic, so they are not suitable for describing the really traffic process. This paper focus on self-similar traffic and its influence on network performance, what's more, the reasons of self-similarity will be discussed in chapter 2.We give the mathematic definition of self-similar process and characteristics ,then introduce how to examine the process if it is the self-similar process, i.e. to find out self-similarity parameter-H parameter. Examination methods include V-T, R/S, Periodogram, Whittle and EM based on wavelets, further comparing five examination methods concisely.Self-similar traffic modeling is the key point in this paper. ON/OFF model, FBM and FGN model, FARIMA(fractional autoregressive integrated moving average)model will be introduced at first. The key model is FARIMA, as this model is capable of capturing both the long-range and short-range behavior of a network. The method of building the model and its implementation are given detail in chapter 4. Concretely , including the implementation of fractional-difference arithmetic operators, foundation of accessorial autoregressive model, maximum likelihood estimation of parameters, calculation of the model's order etc. Finally, we analyze prediction procedure based on upper-limit probability for FARMIA model.Finally, the influence of self-similar on network performance will be disscussed. In network performance, queuing performance and cell loss probability are important indicators, therefore, we apply large deviation technique to FARIMA traffic model for queue analysis, giving out the cell loss probability based on ON/OFF traffic.