Classification and comparison of modelling strategies for VBR Video Traffic

In this paper we propose a framework to classify and compare existing video traffic models. The classification is performed according to the statistics which the models intend to fit: first order statistics (PDF or Probability Distribution Function), Short Range Dependence (SRD) or Long Range Dependence (LRD). After a thorough comparison of the models in a queue, using real MPEG traces, we conclude that, for realistic scenarios, a good fit of the PDF is essential while the effects of SRD have a secondary importance and LRD could be neglected. Consequently, special attention is devoted to the election of a standard PDF which can properly describe video traffic burstiness. On the other hand, we also investigate the problem of "starvation " by means of which LRD properties, which have no impact on real-time video services, could affect the quality of other lower priority services which are conveyed on the same network. In particular, we study the quality of real HTTP (HyperText Transfer Protocol) traces being multiplexed on the same link as video traffic. 1