This paper consists of two parts. In the first one, we present a statistical analysis of experimental MPEG video sources and we test some literature models for the performance evaluation of a queueing system loaded with such bit streams. In the second one, we propose a new modeling approach that considers the particular structure of MPEG streams and gives extremely accurate results in a well-defined set of the queueing system parameters.
Our results show that the impact of correlations and the particular structure of MPEG sequences (the GOP) is very significant. As a consequence it does not seem possible to model MPEG sources with stochastic processes based on only one picture type, with “simple” bit rate distribution, and/or whose autocorrelation function is a sum of exponential functions.
The proposed model is validated with simulation results produced using many experimental video sequences (both MPEG1 and MPEG2); the sequences analysed are long enough to allow a significant statistical analysis.