Balancing the Trade-oo between Prefetch and Rewind Buuering in Video-on-demand Services

The eecient delivery of pre-recorded, compressed video plays an important role in a wide variety of emerging multimedia services. However, compressed video exhibits signiicant bursti-ness that complicates the transmission from the video server, transfer across the communication network, and playback at the client site. By capitalizing on buuer space at the client, the server can reduce transmission overheads and enhance playback performance. For example, the server can decrease the peak bandwidth requirement of the outgoing video stream by prefetching frames in advance of each burst. Alternatively, the service could use the buuer space to save previous video frames, in the hope of satisfying VCR operations directly at the client site. This paper investigates this fundamental trade-oo between network bandwidth requirements and eecient VCR functionality in video-on-demand services. In particular, we characterize the relationship between prefetch buuer space and the peak bandwidth requirement. Using these results, we develop a cost model and eecient numerical techniques for balancing the use of prefetch and VCR space in the client playback buuer. Experiments with sixteen compressed, full-length movies demonstrate that this buuer trade-oo is unique for each movie and can be accurately represented with a small number of parameters.