| Compared with traditional Client/Server Model, Peer to Peer model has good scalability, load balance and self-healing performance. Each node in the system is both the provider and consumer of resources, and is responsible for routing and forwarding messages from other nodes. P2P can make full use of nodes' individual resources and share resources effectively, so P2P can realize various applications which are large scale.According to the mechanism to organize nodes, P2P systems can be classified into Unstructured P2P (loosely organized) and Structured P2P (strictly organized). And this dissertation focuses on the latter. Structured P2P employs some rules to define the method of node organization and message routing. When a node receives a message, it deals with the message according to the rules. Every message is to be transported to a unique next hop, and can reach the destination in a few hops. Rules make Structured P2P scalable and popular in practical applications, but also weaken the flexibility. This dissertation presents deeply studies in the availability of P2P (load balance and topology stability) and search algorithms of Structured P2P. The main contributions of this dissertation are as follows:Load Balancing is one of the fairness issues in P2P system. In order to solve it, this dissertation proposes a tree-based load balancing method. In this method, firstly, a measurement metric for load of files which considers the differences of capability of nodes and popularity of resources is put forward. Then some hot resources are replicated and stored in nodes that form a tree topology according to the metric. It's scalable and easy for the nodes to maintain the tree's information. And the load is very dispersive. This method is proved to be effective by simulation.Existing search mechanisms of Structured P2P are not efficient for multiple keywords search. The dissertation proposes a multiple keywords search model based on DHT. Multiple keywords of a resource are stored in one P2P node, and weighted vectors are used to denote the information of the resource. Queries are also represented by weighted vectors. The desired information can be located by matching vectors of the query and the resource. The simulation results show it can improve search accuracy significantly and reduce routing hops effectively.Structured P2P doesn't support fuzzy matching. Information search method of Structured P2P network is to hash the keyword into an ID and gets results according to the ID. This dissertation presents a novel Interest-based CLustering sEmantic Model ICLEM. Firstly, proposed a semantic conceptual space in which peers are clustered according to their interest and form groups. Secondly, developed thesemantic-based content search based on the framework. Finally, a large-scale network is deployed to simulate, and the results show the feasibility, effectiveness and efficiency of the proposed model.This dissertation gives an implementation of service platform based on structured P2P. In the implementation, in order to solve the problems caused by churn, we proposed some improvement schemes from several aspects, such as the organization of routing table, redundancy storage, and the process of queries. The experiment results indicate the effectiveness of the schemes. The platform provides APIs for upper layer P2P application and realizes a lot of functions, such as content distribution, information search, service publishing, event notification, distributed network measurement and routing optimization, etc. The experiences in designing and implementing of the system are of high value to P2P researchers.
|
PDF Full Text Request