Modeling And Prediction For Self-Similar Network Traffic With Heavy Tailness

In the last twenty years, the high-speeding networks have emerged in the every corner of the modern life. The network technologies are deeply changing the way of our lives. Before the over ten years, the experts from Bellcore Lab found firstly the self-similarity in the network traffic. The self-similarity existing in the network traffic effects extensively and profoundly the total domain of network engineering though the physics mechanism of the self-similarity in the all kinds of the network traffic. The modeling and prediction of the self-similar network traffic have been the focus of the network engineering as the foundation for the analysis of network performance, network management and protocol design. In the last few years, heavy tailness namely non-Gaussian, existing widely in the network is a key factor bringing about the long-range dependence of the self-similar network traffic over the broad time scales. The almost existing the models and predictions of self-similar excluded the heavy tailness factor. In order to meet the needs of the National Science Foundation of China"Adaptive Technique Research about Aerial Interface Based on MIMO-OFDM System"(No.60496315) and the National 863 High Technology Research & Development Program"the Coding, Testing, Traffic and Application of the Digital Audio and Video"(No.2002AA119010), the dissertation begin to research modeling and prediction of self-similar network traffic with heavy tailness.After reviewing the self-similarity and other correlative primary conceptions, the dissertation explains in detail the possible cause of the self-similarity generated in the network traffic. And the dissertation analysis roundly the relationship between the self-similar network traffic and queueing performance, congestion control, scheduling algorithm, TCP transmission mechanism, Quality of Service of multimedia traffic et al. In order to illustrate the importance and urgency of the research on the self-similar network traffic, we clarify the profound impact for the development of the network engineering caused by the self-similarity of the network traffic.Heavy tailness is an important reason of the long-range burstiness appearing in the self-similar network traffic. The alpha-stable processes theory is a perfect tool capturing the heavy tailness of the self-similar network traffic because the characteristic parameter of the alpha-stable processes can describe the non-Gaussian of the stable processes. The alpha-stable processes theory has the more advantage than other theories within the domain of self-similar random processes.Firstly, we introduce the primary conceptions and attributes of alpha-stable processes for the modeling and prediction of the self-similar network traffic.Secondly, the dissertation compares the advantage and disadvantage of the several representative existing models of the self-similar network traffic, furthermore presents a new model of the self-similar network traffic. The model is parsimonious in the number of parameters that have direct physical meaning. We adopt the actual trace from Bellcore Laboratory to simulate the new traffic model. The simulating results show that the model is exact and effective.Thirdly, we present four individual predictors including AR (AutoRegressive), MA (Moving Average), FARIMA (Fractional AutoRegressive Integrated Moving Average) with alpha-stable innovation based on the new model. These predictors under the criteria of infinite variance are capable of predicting the self-similar network traffic with heavy tailness effectively. The last predicted results are achieved by mixing the previous individual predicted values in order to enhance the prediction accuracy. The prediction experiment with the actual trace from Bellcore Laboratory and Lawrence Berkeley Laboratory validates the conclusion.Finally, we summarize the new model and predictors of the self-similar network traffic, point out the application perspective of the research results and the future research direction.