Self-Similar Network Traffic Simulation And Performance Analysis

With the accelerated growth of various network services, the development of communication technology has not caught up with the users' demands. For the purpose of further research on network congestion control, bandwidth distribution, and performance evaluation, we need to build new models which can exactly describe the characteristics of network traffic. Self-similarity can make a tremendous impact on the network behaviour. Therefore, it has become the hot spots and difficulties in the deomestic and international academic circles in terms of network performance analysis, evaluation, and its application research with self-similarity.This dissertation does some research on self-similar network traffic by simulation based on deep analysis of the relative theories. Firstly, it makes a concrete introducation on the self-similar network traffic, including the common definitions and its properties as well as analysis of some reasons that can produce observed self-similarity. Secondly, it introduces mainstream models of self-similar network traffic, especially concentrating on ON/OFF model and its key parameters. And Self-similar network traffic is generated by directly aggregating independent ON/OFF sources. Thirdly, the common evaluation methods of Hurst parameter that describes the degree of self-similarity are discussed in detail, including the realization and evaluation of the key algorithms. Based on the self-similar traffic generated by software simulation as well as collected from real network with specific services, Hurst parameter was estimated by different algorithms and several factors affecting it were also discussed under various conditions.Finally, the dissertation makes an analysis and discussion on the factors that can impact the self-similar network performance. Evaluation indexes of network performance are introduced at first and then a preliminary analysis is made about the relationship between the queue delay (also packet loss rate) and queue service rate under different values of Hurst parameter.By comparing the values of Hurst parameter, we can judge whether the network traffic is normal or not, and then do a deeper analysis on the causes of abnormal traffic. In a sense, it provides very important theory and application foundations for network security (such as IDS) and the design, analysis and performance evaluation of the next generation communication network systems.