Adapting network video to multi-time scale bandwidth fluctuations

Abstract

In this paper, we examined the issue of adapting network video to bandwidth fluctuations in different time scales. We observe that fluctuations in shared packet switching networks occur in different time scales, from less than a microsecond up to minutes and longer. Sources of bandwidth fluctuations arise from noise in physical devices and communication channels, link layer routing and multiplexing, to connection migrations among heterogeneous networks. Magnitudes of such fluctuations measured by peak-to-average rate ratio vary up to several hundred folds. Since the sources of fluctuations exist in both today's Internet and will again appear in the next generation Internet based on the IETF proposal of the Diff-Serv model, there is no hope that these orders of magnitude fluctuations to go away themselves. Network video applications must face the challenge and adapt to the fluctuations. Our investigation on this issue yields the conclusion that a scalable solution is achievable through layered video coding techniques. This paper presents our finding by first discussing sources of the fluctuations in slow, medium and fast time scales. Layered coding techniques, although they were originally developed for different purposes, can be integrated into a single framework to effectively treat each level of fluctuation. The paper presents this framework and concludes with a mapping between layered video techniques and fluctuation levels.