This research studies how emerging broadband networks can efficiently deliver video on demand. The research assumes that video is variable-bit-rate (VBR) encoded. Particular emphasis will be placed on a design which the P.I.s have recently devised, Join-the-Shortest-Queue (JSQ) Prefetching. Not only does this exciting design give constant-quality playback and almost 100% link utilization, but it also allows for immediate commencement of the video upon user request and near instantaneous response to interactive actions (pause/resume and temporal jumps). This research will examine two distinct end-to-end architectures for video on demand. In the first architecture, there is one video server from which all the videos emanate; the video server is attached to a cable headend or to an ADSL switch. In the second architecture, there are multiple video servers and there is a complex, multi-link network which separates the servers from the clients. The research will study JSQ prefetching for the first architecture; it will study a priority-based scheme which mimics JSQ prefetching for the second architecture. This research shall also develop analytically based call-admission rules for our JSQ prefetching design. It shall investigate how the special structure of MPEG-1 and MPEG-2 encoding can be exploited for designing call admission schemes and selective packet discard schemes at the server. It shall develop an auxiliary payload specification for transporting MPEG 1 and 2 video on demand over real-time protocol (RTP). Finally, an implementation of the JSQ prefetching scheme, which will stream several simultaneous videos to clients, will be developed. The research is described at www.seas.upenn.edu/~ross/jsq.html .
