Bidimensional fluid model for VBR MPEG video traffic

The broadband bandwidth of ATM networks allows to develop new video services. To reduce the resource allocation in these networks, the video services employ compression techniques. The most common used technique is the MPEG algorithm. In order to maintain the image quality the MPEG coders generate a variable bit rate. Usually, a smoothing system is inserted between the video coder and the user network interface. The purpose of this system is to improve the performance of ATM networks. In this way, the traffic delivered to the network maximize the multiplexing gain. To evaluate and design ATM networks a characterization of the supported video traffic is needed. In this sense, this paper presents an analysis of the smoothed MPEG video traffic. Likewise, a new procedure to adjust a binomial Markov Modulated Fluid Process (MMFP) is presented. The goodness-fit-tests reveal some deficiencies on the autocovariance and probability density functions on the model. In order to correct these failures a new bidimensional MMFP model is elaborated. This new model is basically the aggregated process of two ON/OFF minisources types. For this point of view, the model can be dimensioned using each minisource type to capture the short and long range dependencies of the video traffic. The results show a well fit of the above mentioned functions.