With the exciting successes of file sharing applications such as Napster, Gnutella and Kazza, Peer-to-Peer (P2P) system becomes one of the fastest growing applications on Internet. A P2P system can be viewed as a distributed system in which nodes have identical responsibility. It has generated great research interests because it achieves many fascinating technical characteristics such as fully decentralized control, self-autonomous and load balance. However, its appealing properties also bring more difficult problems than the traditional Client/Server applications in the system design and implementation, especially in large-scale environments.; A critical problem in P2P system is to search and locate resources among peers. This task is fulfilled by the routing service. As the most frequently executed operations in Peer-to-Peer (P2P) applications, the efficiency of the routing procedures has great influence on system overall performance. In recent years, Distributed Hash Table (DHT) based routing algorithms have been proposed as a fundamental building block for P2P systems. In current DHT algorithms, routing tasks are uniformly distributed across all the system peers. Each peer only maintains a small amount of system routing information and a routing procedure is accomplished by the coordination of several peers. However, the neglect of the characteristics of real P2P systems reduces their efficiency and makes them difficult to achieve the ideal performance. Caching is another important technical aspect. Good caching scheme can significantly decrease the unnecessary network traffic and reduce the client perceived latency.; The main contribution of this dissertation is: by taking the P2P system characteristics such as topology information, peers' computer resource diversity and workload distribution into consideration, system routing and caching performance in current DHT algorithms are greatly improved. We also address maintenance overhead problem and propose our prelimilary solution to reduce it.
展开▼
