Self-Similar Traffic Synthesizing Using Gaussian Mixture Model in Wavelet Domain

It has been recognized that self-similarity of the Internet traffic significantly affects the performance of networks, and traffic modeling and generation is a primary step of network performance evaluation. An algorithm for self-similar traffic modeling and generation based on a mixture Gaussian model in wavelet domain is proposed in this paper. The approximate Karhunen-Lo｀eve transformation inherence endows wavelet with the power of decorrelating long-range dependence, and the mixture Gaussian model exactly captures the non-Gaussian distribution of wavelet coefficients. Both statistical analysis and queueing performance simulation are conducted to evaluate the proposed method. Numerical results suggest that this method can model and synthesize actual network traffic more accurately and has the advantages of low computation complexity of traffic generation in particular.