Unstructured Peer-to-peer Network Routing Optimization Technology Research

In unstructured P2P systems, it is hard to find the desired files without distributing queries widely, since there is neither a centralized directory nor any precise control over the network topology or file placement. Flooding-based routing and search approach is widely used in unstructured P2P systems. In blind flooding, each peer sends query message to all of its neighbor peers, which in turn continuously forwarded to all neighbors. The query is executed hop-to-hop through the unstructured network until success, failure or timeout. By flooding the whole P2P network, many messages are generated. It is shown that more than 70% of the generated messages are redundant for a flooding with a fixed TTL (Time-to-Live) in a moderately connected network. These redundant messages consume bandwidth and waste processing power, but it is useless to enlarge the search range and improve the searching coverage. To address the above problems, many researchers have been studying how to reduce flooding radius while maintaining the satisfactory search quality.In this dissertation, we propose a new routing and search algorithm, Adaptive Flooding Routing Algorithm (AFRA). AFRA combines depth adjustment and width adjustment. The depth adjustment is a method that reduces the number of routing messages by adjusting TTL value. The width adjustment means that a peer does not relay a query to all neighbors, but define the number of neighbors before forwarding the query. To our knowledge, although the existing documents have mentioned width adjustment and depth adjustment, there is no practical solution on how to combine width and depth adjustment to optimize search mechanisms. Experimental results show that our AFRA scheme can reduce flooding message by about 65% while maintaining the acceptable high search quality.To address the severe performance limitation of flooding search mechanism, we propose an Active Recommendation Mechanism Based Distributed Approximate Indexing (ARMBDAI) method. Our scheme can work in high dynamic P2P networks and enable users to share valuable information with others. By Active Recommendation Mechanism (ARM), we build a two-tier proxy architecture consisting of reputation evaluating proxies and content indexing proxies. Based on ARM and the new architecture, we purpose a Distributed Approximate Indexing (DAI) method to support efficient search without the need for a single centralized index host. The effectiveness of DAI is demonstrated through simulation studies. Our experimental results show that DAI solution can improve search efficiency and optimize network topology.