The Research Of Data Scheduling Mechamisms In P2P Streaming System

With the widespread deployment of broadband access, multimedia services are getting more and more popular, and have became a significant application in today's Internet. Multimedia transmission needs the support of broadband access, due to poor scalability, traditional Client/Server model is not appropriate when facing growing users of multimedia services. Comparing with Client/Server model, in P2P each peer acts both as server and client. The whole system could better utilize resources from all over the network, thus greatly improve its scalability. P2P technology provides a new solution for large-scale multimedia services.In this thesis we first introduce the background of P2P technology, features and related research issues. Although the decentralized feature of P2P greatly improve system scalability, it also brings about some challenging problems, such as peer dynamics, data scheduling mechanism, fairness and security issues. The second chapter discusses related work in P2P area, including data coding, network measurement and incentive mechanisms. The third chapter introduces one research direction: how to design an efficient data scheduling algorithm to meet the needs of multimedia transmission. Then we put data scheduling algorithms into specific P2P streaming systems to analyze their cons and pros.In chapter four we proposed a distributed priority-based data scheduling algorithm called DPC. DPC is essentially a pull-based mechanism. It contains two parts: supplier side scheduler and receiver side scheduler. Each peer and data chunk is labeled with priority. Each supplier would choose peers to serve based on their priority, and each receiver would request data chunks in the same way. These strict rules guarantee the geometric distribution of data chunks, and thus minimize the delay for data distribution. However, the strict priority also introduces much message overhead. After analyzing the message transmission delay, we further propose a push-based DPC algorithm. The revised algorithm is pull-based scheme. The suppliers take full responsibility for data chunk distribution, and thus there is much less message exchange between peers.Unlike live streaming, VCR operation and playback asynchrony in P2P VOD systems greatly dilute peers'ability to help each other and offload the server. In chapter five we first discuss the major causes of server load in P2P VOD system, and then propose a predicted-based bandwidth allocation algorithm called PBA. The basic idea of PBA is that data scheduling can be equally treated as bandwidth allocation. Each peer allocates extra bandwidth to help descent peers pre-fetch data chunks. A stable peer with higher playback position is able to obtain a larger share of extra bandwidth. In this way, the differences in the amount of data chunks downloaded by different peers would decrease, and thus reduce the server load caused by departure misses.