Research On Data Searching And Scheduling In P2P Video-on-Demand Systems

In recent years, Video-on-Demand(VoD) streaming services have emerged and attract-ed large amounts of Internet users. However, as the "Killer Application" on the Inter-net, streaming service causes significant bandwidth consumption. Traditional Clien-t/Server architecture based streaming system lacks of scalability. P2P (peer-to-peer) technology has emerged as a promising solution to address the scalability issue. In a P2P based VoD system, the users cache the video content received, and relay the cached data to other users. This cache-and-relay strategy alleviates the server stress and increases the scalability of the system. However, due to the asynchrony of the user requests and dynamics of the interactive operations, there may be differences among the data downloaded and cached by the users. An efficient data searching approach is required for the data transferring corporation among the users, by which a user can find other users holding the expected video content. The realtime requirement of the streaming service also desires an efficient data scheduling scheme, to make sure the video data can be downloaded in time.Based on the characteristics analysis of P2P VoD systems. We aim to design efficient data searching and scheduling schemes to improve the performance of the systems. The main contributions of this thesis are as follows:1. We looked into the data scheduling problem in P2P VoD systems with central-ized indexing capability. We first analyzed the characteristics of the data trans-ferring in P2P VoD systems, and discussed the objectives of the data scheduling. Based on the analysis and discussion, we have proposed the max-flow based data scheduling algorithm MFDS. MFDS maps the data scheduling problem onto a virtual flow network. We solved the data scheduling optimization problem by computing the max-flow of the mapped virtual flow network. Theoretical analy-sis and simulation verified the good performance of MFDS, the low computation cost and low node degrees of the scheduling result.2. We looked into the distributed data searching problem in P2P VoD systems. We have proposed a Skip Graph based indexing overlay for P2P VoD systems named SkipStream. In SkipStream, users are gathered into several clusters according to their playback point. Based on their keys, these clusters are organized into a special skip graph. We designed a distributed data searching scheme on this skip graph based overlay. Theoretical analysis and simulation have shown the effi-ciency of the searching scheme, as well as the robustness and low maintenance cost of the search structure.3. We looked into the distributed data scheduling problem in P2P VoD system-s. We first analyzed the essence of the distributed data scheduling in P2P VoD systems. We analyzed and verified the efficiency of the Early Reject combined EDF(Earliest Deadline First) strategy for the requested peers. We also analyzed existing representative strategies for the requesting peers by simulations. Based on the analysis, we have introduced the concept of upload pressure vector, and have proposed the upload pressure vector based data scheduling scheme. This novel scheme determines how to send out the data requests and how to deal with the received data requests for each peer. Simulation results have shown that the upload pressure vector based scheduling scheme outperforms the existing ones.