Time Series Methods for Synthetic Video Traffic

The scope of this paper is the creation of synthetic video traffic using time series models. Firstly, we discuss the procedure for creating video traffic with FARIMA models. However, the created traffic displays the LRD characteristic of real traffic, but underestimates its moments (mean, sd, skewness and kurtosis). We present two approaches for improving the popular FARIMA model for the creation of synthetic traffic. The first approach is to apply FARIMA models with heavy-tailed errors for traffic creation. The second is a two step procedure, where we build a FARIMA model with normal innovations and then we provide a statistical transformation for its projection in order to catch a desired marginal probability distribution. Using this procedure we approximated Student t and LogNormal as marginal distributions. The above procedures are applied to the performance evaluation of three real VBR traces.