Analysis And Modeling Of Network Traffic Based On Wavelet Transform

A great deal of research results indicate that actual network traffic processes exhibit ubiquitous properties of multi-scales, namely self-similarity (mono-fractal) in large time scale and multi-fractal in small time scale. Multi-fractal offers a good mathematical framework to describe the singularity of traffic in small time scale. It is necessary to use wavelet transform to study the multi-fractal because of the decorrelation characteristic of Wavelet. The multi-scale characteristic of network traffic also needs to introduce new methods to study its essential characteristic.The multi-fractal characteristic of traffic has great effect on the performance of network. It is necessary to construct traffic model based on multi-fractal characteristic. Firstly, with the analysis of the characteristics of whole and local scale of actual network traffic, this paper ascertains the fractal characteristic and its initial time of traffic, and compares the different performance of traffic in different time scale, and analyzes the reason for the occurrence of this phenomena. In order to deeply study the factors which affect the characteristic of actual traffic under the condition of multi-fractal, the author using wavelet transform to analyze the dependence of the actual traffic, he divides the traffic into blocks and then internally or externally shuffles them. The test results show that mean and variance of the traffic both have great impact on multi-fractal.Secondly, based on the multi-fractal characteristic of the traffic and the property of decorrelation of wavelet transform, a novel synthetic model is constructed to improve the synthetic accuracy. Through the scale description and performance assessment of the synthetic traffic, the validation of the novel model is verified.Finally, the necessary of the construction of network traffic model based on multi-fractal characteristic is discussed. The model can predict long-range and short-range dependence of the actual traffic. Based on the characteristic that AR and ARMA models can correctly predict short-range dependence while has low prediction accuracy to long-range dependence, a novel prediction model is constructed based on the superiority of decorrelation of wavelet transform. This novel model has also high prediction accuracy to long-range dependence. At the same time, the improved model also conquers the defect of computing complication of FARIMA model and keeps the briefness of algorithm.