This thesis studies distribution schemes for stored video over broadcast networks and over the Internet. In the first part of the thesis we consider designing near Video on Demand (VoD) systems that minimize start-up latency while maintaining high-image quality. We propose a novel scheme for near VoD which is based on periodic broadcasting of VBR-encoded video, and develop a series of multiplexing schemes for broadcasting VBR video that are based on smoothing, buffering and client prefetching. Our multiplexing schemes achieve high link utilization with negligible packet loss, and result in significant reductions in the start-up latency relative to the CBR case. In the second part of the thesis, we consider the design of stored video streaming applications for the current Internet. We first develop a theoretical formulation and analysis for the streaming of layered-encoded video. We suppose that the video is encoded into two layers and model the bandwidth available to the application as a stochastic process whose statistical characteristics are unknown a priori. We address the problem of dynamically allocating the available bandwidth among the two layers in order to minimize the impact of client starvation on the quality of the decoded video stream. We next consider streaming stored video over actual fair-share bandwidth conditions. We perform a series of Internet experiments in which we collect TCP throughput traces. We find that fair-share bandwidth varies over a broad range of time scales. We investigate the impact of the long-term behavior of fair-share bandwidth on several data transmission schemes for streaming video. We quantify the performance gains achieved with prefetching non-layered video into client buffers during playback. For layered video, we propose threshold-based inter-layer bandwidth allocation schemes, and develop a heuristic for implementing such schemes. Finally, the thesis presents a comparison study of two rate adaptation techniques for video streaming, namely, layered streaming and switching among multiple encoded versions of the same video object.
展开▼