Research On Decreasing Probability Flooding Algorithm In P2P Network

With its nature of peer-to-peer,P2P (Peer-to-Peer) network can be free from the performance bottleneck of client/server model and therefore has been gaining attention increasingly in the research area recently.In P2P network,one of the key problems is how to search and locate the resources wanted. Gnutella network or its variant of Gnutella-like network, is an important category of P2P network and has mainstreamed most of the existing P2P applications in the internet nowadays because of its characteristic of pure decentralization.However,the flooding-based resource-searching method adopted in Gnutella network always consumes too much bandwidth and thus makes the bad scalability a large hurdle ahead of it.In terms of this,DPFSL(Decreasing Probability Flooding algorithm combined with Self-Learning) is proposed in this thesis.It is a new simple and efficient algorithm aimed at improving the scalability of Gnutella network,combining two features of decreasing probability and self-learning. The core ideas behind DPFSL are :(1)the probability of flooding each peer in the network decreases with the increase of the distance between the peer and the query-initiating peer.(2) each peer in the network has the ability to learn related resource information from query responses.In order to do this,each peer has an ET which is a table used to record resource information.When a query response goes from a query-responding peer to the query-initiating peer,each peer forwarding this response will:if the resource information contained in the response does not appear in the ET,then add it to ET;otherwise, if the resource information is obsolete,then update it.In comparison with the flooding method,the three noticeable improvements of DPFSL are:(1)flooding will stop when a query is met,with no regard to the non-zero TTL field in the query.(2)if a query can not be met,the receiving peer will forward it to all of its neighbors independently with a corresponding probability in contrast with probability 1 adopted by flooding method invariantly.(3)self-learning feature which can avoid the blindness of flooding counterpart.(4)good inheritance of simplicity for implementation of the flooding method. Based on the metrics of query coverage percentage,query hit rate and query dissemination percentage,DPFSL has been evaluated through simulation test.The results show that while maintaining a high level of query hit rate,DPFSL can satisfactorily decrease the query coverage and dissemination percentage,which has proven it a promising candidate for improving the scalability of the Gnutella network.