Characteristics Analysis And Performance Evaluation Of Network Traffic

The Fractal characteristic of Network traffic has impacted on the network performance in many aspects, such as network design, traffic control, network plan, network management and QoS. Differ from traditional traffic modeling(e.g. Poisson modeling), one of the critical characteristic of modern network traffic is self-Similarity, that is network traffic has the characteristic of statistical similarity (high correlation) and heavy tailness (high variation or burstiness) in a wide range of time scale. It shows that the correlation structure of network traffic changes with the tendency of slow attenuation. Compared with self-similarity, multi-fractal of aggregated network traffic describes the burstiness under multi-scale and the variation of burstiness with time, which is one of the most significantly statistical characteristics of the network traffic.Self-similarity and multi-fractal have been studied in this thesis. Firstly, the Hurst index which describes the degree of the self-similarity is estimated, and its accuracy,reliability and limitation are studied and analyzed through simulations.Secondly, based on the induction and summarization of the basic concept of self-similarity, the realization process of the common self-similar models such as ON/OFF,FBM/FGN,FARIMA are studied, and the accuracy of self-similar flow sequences generated by these models are analyzed.Thirdly, the parameters of the multi-fractal traffic, namely scaling function and the moment factor are estimated and the structure of the multi-fractal traffic is tested.Finally, based on the analysis and the parameter estimation of the practical data, the traffic is generated based on JMF model. The traffic data generated by JMF model and FGN model are used to drive the OPNET to perform the performance simulation.The results of the performance analysis and adaptivity study of different traffic models shows that the characteristics of the practical data can be discriminated through the parameter estimation. Compared with FGN model, the theoretical flow generated by multi-fractal model is more close to the practical traffic with the same parameters, and its simulation results are more accurate. It verifies the multi-fractal characteristic of the network traffic from another aspect.