| Summary: | 碩士 === 輔仁大學 === 資訊工程學系碩士班 === 107 === In recent years, the rapid growth of network multimedia audio and video demand has placed a big burden on traditional video servers in the client-server architecture. The peer-to-peer video on demand architecture has therefore become popular. The peer-to-peer multimedia service splits the video into small segments, each of which can be downloaded from different nodes. Each node needs to retain some video pieces in order to serve other neighbors. However, it is a node retains the playbacked usually impossible for a node to store all the fragments due to its limited buffer space. If the pieces in the buffer randomly, its neighbors may not be able to download the required pieces from it, and they have to download from the video server instead, thus increasing the server load.
In order to make more efficient use of the peer-to-peer architecture, we need an intelligent way to select the video pieces to be refrained in the buffer. In this paper, we propose a mechanism for managing the buffer space of a node by evaluating the resources of its neighbors. Specifically, for each piece in the playbacked area, we compute its potential value to the neighbors, and choose to retain the pieces with higher values. The experimental results show that compared to FIFO, our proposed method can reduce the load of the video server by 48%, and reduce the number of pauses by 26%.
|
|---|
