| Summary: | 碩士 === 國立清華大學 === 資訊工程學系 === 95 === Abstract
Many video-on-demand (VOD) systems have adopted a peer-to-peer (P2P) approach to deliver the video streams. By leveraging the resources in the participating peers, the P2P VOD systems can alleviate the severe resource demands at the video servers. A typical use of the peer resources in P2P VOD systems is to let the peers buffer the video streams after viewing so that other peers can download the streams from those buffers instead of from the servers. Most previous works assume a fixed buffer size for simplicity of implementation. When the buffer is full, old video segments are just discarded. The problem is that the peer may miss the chance of serving another peer if the latter happens to behind the former by a buffer-full of video. If the buffer size can be extended dynamically, a peer may serve more peers and the server load may be further reduced. However, extending the buffer will likely incur virtual memory overhead and degrade the performance of the local computer. Apparently, the idea of dynamic buffering must be applied carefully. In this thesis, we study the various issues of dynamic buffering for P2P VOD systems. We first introduce a realistic model of system performance under different streaming buffer size. We then propose a system DBS-Net to adjust the available buffers according to the condition of peers so that the memory space could be better utilized. Simulation shows that DBS-Net outperforms the systems with a fixed buffer size.
|
|---|
