Modeling VBR video traffic by Markov-modulated self-similar processes

It is estimated that video traffic will increasingly occupy a major portion of future network bandwidth, and thus traffic modeling plays an important role for network design and management. In this paper, we propose Markov modulated self-similar processes to model MPEG video sequences that can capture the LRD (long range dependency) characteristics of video ACF (auto-correlation function). The basic idea behind this modeling is to decompose an MPEG compressed video sequence into three parts according to different motion/change complexity. Each part can individually be described by a self-similar process. In addition, beta distribution is used to characterize the marginal cumulative distribution (CDF) of the video traffic. To model the whole data set, a Markov chain is used as a dominating process to govern the transitions among these three self-similar processes. Initial simulations on a real MPEG compressed movie sequence of Star Wars have demonstrated that our new model can capture the LRD of ACF and the marginal CDF very well. Video traffic synthesis using our model is presented. Further research in this direction is discussed