The Research Of Self-similar Traffic And Simulation Based On NS2

With further research on network and the development of measurement techniques, Lots of measurements done on actual network show that the real network traffic presents self-similarity property. The Study of the characteristics of network traffic is basic method and way to deeply understand the essence of network, and to comprehend the operation of network. It is also the important approach to improve the performance of network, to optimize the design of network and to implement network engineering. Through network measurement, simulation and theory analysis, this dissertation studies the self-similarity property of network traffic.Firstly, fundamental theory of scale conduct on traffic, self-similarity property, and LRD, heavy-tailed distribution is narrated. We propose several common analytical methods of self-similarity property and summarize such issues as self-similar traffic modeling, cause and its impact on network performance. Secondly, through ON/OFF model respectively submits to the four conditions of the different distributions, we analyze and illuminate that the influence of heavy-tailed distribution on self-similarity property.At the same time, for Weibull distribution, it proves that Self-similarity originates from the property of traffic sources other than the mere aggregation by means of both theoretical analysis and computer simulation. The thesis makes an implementation of the self-similar traffic which is completed by the aggregation of Weibull distributed ON/OFF sources based on NS2.Thirdly, adopting FGN model, through the comparison of the simulation data, we analyze the aggregated characteristic of self-similar traffic. We conclude that Hurst parameter of the aggregated traffic isn't influenced by the values of mean of the individual traffic under certain circumstances. Under the error circumstance, it is approximately the H value of the individual traffic that owns larger variance. Moreover, we carry out another simulation experiment for self-similar traffic and discover that the tail of the traffic ratio of the two successive arrivals is close to Weibull distribution. This makes sense in the field of self-similar traffic prediction.Finally, for the alterable grouping length network, we make a further simulation analysis on the effect of self-similar traffic on the performance of the GI/GI/1/K/FIFO queuing. Thereinto, the part conclusions are in harmony with the analyses of ATM queuing, whose grouping length is constant. In addition, we complement the simulation conclusions of ATM queuing performance from other sides.