Research On Self-Similarity And Bursty Network Traffic Thoery And Network Traffic Model Analysis

Recently, with kinds of mass multi-media applications being carried over the network, theories of network traffic model and researches on network traffic analysis have caught researchers'attentions in the fields of computer network and communication. With the support of project of National Nature Science Foundation of China(NSFC),"Based on Covariation-orthogonality and Combined Optimization multimedia network performance prediction model (60502023)"and project of"experiment research of access network topology of MAN data service"carried out by the department of electronics and information engineering of Huazhong University of Science and Technology(HUST) and Zhuhai mobile company, this dissertation is focused on modeling and analysising the network traffic. The key problems in this research field, such as the network traffic modeling, parameter estimation and model analysis are studied in this dissertation."Self-similarity"is a keyword in current network traffic model researches. Compared with telephone traffic, it is main characteristic of multi-media network traffic. Self-similar models are more suitable for current network traffic because the network traffic characteristics can be present better by these self-similar models. Thus network traffic research based on self-similarity theory can catch network traffic essences better.First of all, kinds of researches on network traffic in real network environments are analysised and self-similarity theory is introduced generally. Then research on network traffic by self-similarity theory is decided, which is focused on self-similarity definitions and Hurst parameter estimation methods. At the end of introduction, based on theory research and network traffic characteristics, burstiness, long correlation and self-similarity, Alpha-stable distribution and corresponding self-similar stochastic processes is thought of as theory foundation. Subsequently, applications of Alpha-stable distribution in traffic model and network resource scheduling fields are studied.To be more accurate on analyzing network traffic characteristics, definitions and PDF of Alpha-stable distribution and properties of Alpha-stable distribution are analysised and studied. Especially, problems, such as how to get expression for the PDF of Alpha-stable distribution and how to get random variables conforming to Alpha-stable distribution, are studied. These researches on Alpha-stable distribution are key for network traffic simulation and bursty network traffic research.Researches on Alpha-stable distribution theory and properties in sections above are for theory analysis. On the other hand, researches on characteristic of network traffic need their distributions, which are gotten by parameter estimation methoeds.Thus Alpha-stable distribution parameter estimation is studied respectively. There are many methods for parameter estimation of Alpha-stable distribution, each of which comes from different theory. Here, primary parameter estimation methods of Alpha-stable distribution are studied and classified into five types: Characteristic Function Method, Quantile Method, Maximum Likelihood Method, Extreme Value Method, Moment Method. By analysis and comparison, each type method is studied particularly. Two problems neglected in the past researches, the impact of sample numbers to maximum likelihood method based on FFT and approximation accuracy of extreme value method, have been gotten.With the conclusions above, based on analysis of Alpha-stable distribution and Alpha-stable distribution parameter estimation methods, a new Alpha-stable distribution parameter estimation method based on extreme value theory is advanced. The new method makes use of truncated extreme value and classifying skewness to improve performance. Simulation results show this new method is more precise and stable.Futher more, analysis of Alpha-stable distribution is used for analyzing self-similar network traffic model based on stable increasement. Because of the burstiness of network traffic, network traffic is analyzed based on linear fractional stable process whose increasement is stable. This new method without any hypothesis is a general method. Simulations show this new method could work well.In the last section, main research conclusions are summed up and listed. Also, The some new directions and further problems in network traffic research fields are given out and analyzed.