Multifractal Modeling And Prediction Of IP Network Traffic

With the rapid development of the IP network, modeling and controling of network traffic become more and more important. Since encoded video traffic is expected to account for a large portion of the traffic in future wire-line and wireless networks, a clear understanding of video traffic becomes a basic task in traffic engineering. This dissertation focuses on modeling and analysis of video traffic and network traffic. The main results and contributions of this dissertation are as follows:A new frame level MPEG-4 VBR traffic model base on Spatial Renewal Process (SRP) is proposed. SRP can describe its autocorrelation function and marginal distribution separately. This feature is utilized to set up an SRP-based MPEG-4 video traffic model. One background process is used to describe the scene length distribution, and the other is used to describe the marginal distribution of frame size. SRP-based model can depict the long-range dependence (LRD) and packet loss rate accurately than traditional Short-range dependence SRD models. Moreover, it is simple and easy to be analyzed.A novel multifractal model (PMFM) is proposed in the paper. Existing multifractal models always use statistics match of multipliers to control model's LRD. Unlike previous studies, the correlation function and marginal distribution of multipliers of multifractal multiplicative process are investigated. SRD model of root approximation coefficients is used to control LRD of final sequence. By modeling the coarsest coefficients with the auto-regressive (AR) process, PMFM can be used for traffic prediction at large time scale while holding the multifractal nature of original traffic. Furthermore, we choose the distribution of multipliers at each time scale by Kullback-Leibler (K-L) distance. Simulation shows that our model has better accuracy and stationarity than traditional MFM.In contrast with single step prediction, it is more significant and useful to predict VBR video traffic by multi-step at large time scale. However, together with the slowly-decaying auto-covariance function (ACF) and the traffic non-stationarity, it suggests that some conventional prediction tools, which only use SRD feature, such as the linear auto-regressive method are not appropriate for VBR video traffic prediction. In this paper, we use multifractal method to analysis VBR video traffic and find a useful property. By remaining the curve shape of ACF unchanged, we can convert the original LRD trace to a series of SRD sequence in multifractal domain. Based on it we proposed a multi-step prediction method. Because the LRD feature of trace is used, the multi-step performance of proposed method is much better than traditional methods.