Modeling And Performance Analysis Of Network Traffic Based On α-Stable Self-Similar Process

In the information age of 21 century, with the popularization of network application and growth of the network user, the network has become an indispensable part of human activity. So, the research and analysis of the characteristic of the network become more urgent. For a long time, traffic modeling and analysis is based on Poisson distribution and Markov's theory. But recently measures of network traffic have shown that packet/cell traffic through telecommunication networks exhibits long-range dependence and self-similarity, and this kind of characteristic can't be described by traditional model. Therefore, people have proposed some self-similar traffic models to portray the characteristic of the network, the most frequently used one is Fractional Brownian Motion (FBM) process model. Further studies have also indicated that besides self-similarity and long-range dependence, network traffic exhibits more complicated characteristic, such as strong burstiness. For FBM model can't capture this characteristic, we need to study the new model that can capture long-range dependence and strong burstiness as well.Firstly, this thesis carries on research on the self-similarity characteristic of network traffic, the impact on performance of the network of the self-similarity, the estimation method of the self-similarity coefficient, and the analysis of the existing modeling method of traffic.Secondly, this thesis points out the problem of FBM model on the basis of further investigating the model. We verify the non-Gaussianity of network traffic and propose the use of alpha stable distribution in traffic modeling. Based on the research on alpha stable distribution, this paper introduces linear fractional stable noise (LFSN) process model which can capture long-range dependence and strong burstiness. This paper provides the estimation methods of the 4 parameters of this model and the algorithm producing the flow. Moreover, the advantage of the model is verified by modeling real traffic data. By using FBM and LFSN traffic model to fit the classical Bellcore network traffic, we find that the LFSN model can reflect the burstiness more effectively. Based on the available buffer overflow probability of LFSN model, the formulas for computing the average queue length, the variance of queue length, average delay and the variance of the delay are derived.Finally, the variance of the system performance indices is studied through theoretical analysis and simulation. The results show that stronger burstiness or higher the link utilization will cause the worse of packet loss probability, average delay and jitter. The influence of Hurst index is much more complicated, the impact on performance of network is different in different buffer or speed. We give detailed research on various kinds of situations and studied the impact on performance of the network of the scale effect. Through setting up the simple scene model, the improved reason for the performance of the network caused by the scale effect is analyzed.