Index Replication Strategies In Unstructured Peer-to-Peer Networks

In the peer-to-peer (P2P) network, each computer can act as a client or server for the other computers. It doesn’t cost too much overhead to maintenance the topology when the node joins the network or leaves the network in the unstructured P2P network. So the unstructured P2P network will be used while nodes joining or leaving network is frequent.Resource location is the hot issues in the unstructured P2P network research field. Each node doesn’t have the topology of the whole network. The existing search algorithms originate from the flooding algorithm, and perform more efficient than it. The hit rate of resource location, number of the network messages, network latency and network load are the criteria to measure the performance of the resource location algorithms.Content replication in P2P network generally provides decrease of packet hops until a search succeeds, and improvement of search success rate. It also provides decrease of network load, and realizes efficient search. It is important to decide the number and the placements of replicas how many and where the replicas are allocated. We do some research work based on replication strategies as below.Firstly, we propose an index replication instead of content replication and design two replication methods which called index owner replication (IOR) and index path replication (IPR). In the IOR strategy, the content indices are also allocated when the replica is allocated on the requester. In the IPR strategy, the indices are allocated on all the nodes on the search path from the requester to the holder. Resource is searched by the indices, not by the content. The simulation experiments shows that IOR and IPR perform better than OR and PR on the hit rate and the network latency when there is no storage limitation and no nodes joining or leaving the network.Secondly, we do some research on the index consistency, and propose an index update scheme. There is two cases cause of the index inconsistent. One is the node leaving the network. The other is the replicas replacement. The node sends the indices update messages to the other nodes which the indices point to, and the other nodes forwarding the update message, too. The simulation experiments shows that IOR and IPR with the index update scheme perform better than OR and PR although there is storage limitation and nodes joining or leaving the network.Lastly, we propose a hybrid resource location strategy. The hit rate will be decline when the indices are not updated completely. In order to solve the problem, content search will be used to resource location after the index search failed. The simulation experiments shows that the hybrid resource location strategy perform better than OR and PR on the hit rate and the network latency when there is storage limitation and nodes joining or leaving the network frequently.