Research On Network Topology Model Of P2P VOD System

Introducing the technology of P2P in VOD system can effectively improve the system's comprehensive service capabilities through mutual cooperation of nodes. There are mainly two kinds of overlay network: tree-based overlay and mesh-based overly. The tree-based overlay can future divided into single-tree and multi-tree. The single-tree is efficient, but it is difficult to maintain continuous playback because of the node's dynamic departure and joining. The multi-tree which usually combines video data encodings strengthens the robustness of the system, but it also brings more maintenance overhead. The advantages of mesh-based overlay are robust and stability, however, it is not suitable for the dispersed data request of VOD system. Both overlays fail to meet the real-time demand of VCR operation.Aiming at the oversized latency of VCR operation and poor system scalability in existing VOD system, a new VOD system model, named CS-VOD, with high stability and good expansibility is proposed by using clustering and partitioning strategy. The characteristics of this system are shown as follows.1) The nodes in the system join in different section which represent a fixed region of presentation time according to the time of request data. The system selects section server nodes in each section according to the buffer map of nodes to server the nodes in corresponding section, which can effectively decrease the server load. 2) Each node maintains the nodes whose playing time is contiguous in its cluster. It can fast recover when the partner node is failure. 3) Each node maintains part of nodes in the next cluster and the next section to perform a high efficiently jumps operation. 4) The server merges or segments some of sections periodically based on the number of section servers, thus enhances the utilization ratio of nodes'data buffer and balances the load of the system.Based on the user behavior analysis, a self-adaptive data prefetching strategy is proposed. This strategy fully integrates the advantage of existing strategies, firstly using sequential download to resolve the pause problem caused by single-point failure, then prefetching rare data and anchor-based data to increase the redundancy of rare data minimize the delay time of jump operation. The result of experiment shows that using the self-adaptive data prefetching strategy can adapt to the heterogeneous nature of network preferably, effectively utilize the bandwidth of nodes and decrease the server load.