Workload models of VBR video traffic and their use in resource allocation policies

The load generated by new types of communications services related to multi media and video transmission is becoming one of the major sources of traffi c in WAN networks. Modeling this type of load is a prerequisite for any per formance study, In this paper, we approach the load-characterization proble m from a global point of view by analyzing a set of 20 video streams. We de veloped resource-, subject-, and scene-oriented characterizations of coded video streams. We have also implemented multidimensional data-analysis tech niques and applied a "scene working set" approach, as well as static and dy namic video traffic models. We show that the behavior of a video sequence c an be predicted with high accuracy by applying the scene working set techni que. Applications of this technique for predicting bandwidth demands and al locating buffers in an ATM switch are also described in this study.