Network Traffic Self-similar Model Of Analysis And Research

Since Leland, Taqqu, Willinger and Wilson pointed out the existence of self-similarity in the network traffic, researchers world wide measured and analyzed the network on the run, the result of that demonstrated whatever topology is, statistics behavior of self-similarity could be found.Traditional model based on Poisson arrival process can not characterize the behavior of real network traffic. It is quite necessary to establish new model to reflect the characteristics of self-similar traffic.This thesis firstly introduces what is the so called self-similarity, namely long range dependence, its mathematical definition and statistics characters, and the influence the self-similarity of network traffic bring to network performance. After that, it discusses the method how to detect self-similarity in a given process series, the way to estimate its Hurst parameter. Finally it shows that there does exist self-similarity in the real network traffic through an example.Self-similar model often used nowadays is deeply studied and summarization is given, including the merits and demerits of the models, algorithm complexity, and the best way the models adapting to use. We do all this through extensive simulation which shows there is dependence between precision and complexity of the model, namely the more precision the model requires, the more complex the model is. Major work of this article is to develop a new self-similar network traffic model based on ON/OFF model and chaotic map model. The model simplifies the mapping function of chaotic map model and it makes the state sojourn time still have a heavy-tail distribution with random parameters involved. the idea why to construct this model and characterization of statistical behavior of the model through mathematical analysis is introduced. Then it establishes the connections between the model parameter and the self-similar parameter through ON/OFF model theory, which makes the model under control and capable of capturing different real network traffic. Simulation result demonstrates that the estimated Hurst parameter do accord with the theoretical one. Finally the statistical hehavior of simple queuing system with the traffic generated by the model has been deduced which provides a theoretical basis for queuing analysis.The model brought forward in this thesis could characterize the statistical behavior of network traffic to some extent and it provides an solid foundation for construction, analysis and evaluation of the network.