Cluster-based smoothing for MPEG-based video-on-demand systems

This paper proposes a new technique called cluster-based smoothing to reduce the average per-stream effective bandwidth for the transmission of MPEG compressed video streams. By rearranging the transmission order of frames within windows, the technique exploits the periodic structure of an MPEG video stream, and the bit rate fluctuations across scenes. Trace-driven simulation with empirical MPEG video traces is used to demonstrate the performance advantages of the new technique. The results show that clustering based on frame types in an MPEG video can significantly reduce the per-stream effective bandwidth, particularly when clustering is combined with a modest level of inter-frame smoothing at the VOD source. The buffering and delay requirements associated with cluster-based smoothing are also quantified, and found to be reasonable.