Research Of Reputation Management In P2P Networks

Peer-to-Peer (P2P) network is dynamic, open and anonymous network. All of the peers are equal in the network, which is different from Client/Server structure, and there is no authority center to manage the peers. The rate of successful transaction can be improved through reputation management established on the networks.The main role of this thesis is to establish reputation management system in P2P networks. There are some problems to solve:without the trusted authority center from the third party. The first problem is where to store the reputation information to maintain the integrity and reliability, which are convenient for other peers to query. Second, whether is the query efficiency of the reputation information correlative to the system structure? Third, how to balance the relationship of network load and the number of the query messages. Forth, how to deal with the ratings of the object peers obtained from other peers to compute and evaluate their reputation. The thesis gives a detail analysis of the reputation management from the points of storage, query, propagation, computing and evaluation, and the main contributes are as follows.(1) Two schemes are proposed to manage the storage of the reputation information:the first scheme:a local storage of reputation management scheme is proposed, and it doesn't need the third party to validate the integrity. The reputation is stored in the certificates of the peers and maintained by themselves, and it's for the transaction partners to check the integrity of reputation certificates through validating the sequence of the transaction IDs. The partners will rate each other and sign the information to prevent it from being tampered by anyone after every transaction. The corresponding protocol realizes the integrity validation and creation of the reputation certificates at the same process. This method can reduce communication overhead and improve query efficiency. It's also resistant to some attacks. The protocol can also be applied to other models with good scalability. The second scheme: an innovative cheat-proof reputation management scheme based on RSA (t, n) threshold secret sharing is proposed. This scheme employs threshold secret sharing, and the dealer divides the reputation information into n secret shadows and distributes them to different participants. The reputation information can be reconstructed by any of t or more participants, and cannot be reconstructed by any fewer than t participants. The dealer also distributes the validation shadow corresponding to the secret shadow to the participants. Any unlawful participant palming himself holding secret shadow or rightful participant offering false secret shadow will be recognized unless he can break through RSA cryptography, so it is a secure and valid scheme.(2) Query efficiency is analyzed in different frameworks and the query performance is stress introduced in the partial decentralized framework which is divided into some domains, and query efficiency is analyzed from the view of inside domain and outside domain. We draw the conclusion that the entities, which have more transactions, divided into a same domain will clearly reduce the system cost and message cost.(3) A model about propagation of reputation is built up. It reveals the relationship of the number of query, connection rate and response rate. The model can explain some phenomenon of eBay, and it is a valuable reference to set the scope of TTL in Gnutella, which will be significant to efficiently solve the problems about bottleneck of bandwidth and denial of service attacks.(4) Computational reputation model is proposed for efficiently rating the reputation of the target peers. This model not only considers the direct transactions, but also queries other peers about their direct transactions with target peers and their query results before transactions. Through putting different weights on these data to get the integrated reputation, the model encourages the good behaviors and punishes the bad ones with encouragement factor and punishment factor. The experiments show that the model can hold back malicious peers and ensure higher successful transaction rate.(5) Evaluation of reputation is proposed based on fuzzy set theory. Feedback data, which are expressed in the different forms, are translated into a uniform form with fuzzy set. The last evaluation result uses the membership function based on fuzzy number. The experiment shows that the data expressed with fuzziness and uncertainty is more accordant with the definition and evaluation of the reputation, and the methods based on fuzzy set are more convenient for the entities to predict the forthcoming transaction behavior of the object entities.