Reseach On Trust Model And Incentive Mechanism In Peer-to-Peer Networks

P2P networks are open, dynamic and heterogeneous, and whose nodes are peer topeer and share their resources voluntarily. In recent years, the advantages of P2Pnetworks have been confirmed in some application fields including file sharing,collaborative computation, distributed storage, multicast of application layer, and soon. However, many facts also indicate that large numbers of self-interest rationalusers appear in P2P network and their basic purpose is just to maximize their owninterests while ignoring the whole utility of the network. More seriously, somemalicious nodes adopt cheating behavior to provide deceitful resources and badservices in order to damage the P2P network, and thus this result in data pollution,free riding, common tragedy and usability of incredible services. These problems notonly lower resources utilization, credibility and success rate of services in P2Pnetworks, but also aggravate the load and congestion in physical networks, whichhinder the healthy development of P2P networks seriously. Focusing on theseproblems this thesis tries to propose some theoretic methods and practical techniquefrom construction of overlay network, trust model and incentive mechanism topromote the effective cooperation between nodes and reasonable use of networkresources, and raise the comprehensive performance of P2P network.Firstly, this thesis proposes a construction method of layered P2P overlaynetwork to raise the matching degree between logical overlay and physical networksand support the subsequent hierarchical computation of global trust value. Themismatching between logical overlay and physical networks can bring about a seriousroad-round problem and result in huge bandwidth waste and system delay, andaggravate the common tragedy. Aiming at the phenomenon, this thesis introduces thesubtractive clustering method based on density in fuzzy logic, and designs analgorithm which divides a large network into some small subnets based on theproximity of distance between nodes in physical network, and can ensure the nodes inthe same logical subnet are close in physical network. Each subnet self-organizes astructured P2P network based on Chord protocol. Considering the ability differencesof nodes in P2P network, at the same time, this thesis designs a super-node electionalgorithm based on DHT where the nodes which have strong compute ability and longon-line time are selected as super-nodes of each subnet. All these super-nodes self-organize a small network named top-level network, and thus the layeredstructured overlay network with super-nodes is constructed. Some analysis andexperimental results show the layered overlay network can solve the road-roundproblem and improve the stability of P2P network.Secondly, this thesis presents a global trust model and its distributedimplementation in P2P networks to help nodes to select services and avoid polluteddata and unreliable services as far as possible. The model computes local trust valuebased on history trade information between nodes and obtains global trust value ofnodes based on the whole network via a distributed iteration algorithm. This modelcan ensure such Markov characteres as nonperiodic and irreducible of the probabilitytransfer matrix constituting the iteration algorithms and thus convergence of theiteration algorithms. The model introduces fuzzy function to process positive andnegative evaluation in the local trust computation. Comparing the existing methodswhich add and subtract the positive and negative evaluation simply, this method notonly improves the rationality of local trust computation, but also takes incentivemechanism into account. Focusing on the issue that the computation andcommunication cost of the global iteration algorithm for P2P networks incease withthe square of the network scale, the model in this thesis adopts the hierarchicalcomputation method to obtain the global trust value and divides the large scalephysical network into some small networks logically by the above-mentioned layeredoverlay network method. Iterations are done in these small networks concurrently andthe global trust values of nodes are obtained through computation layer by layer afteriterations have converged. The method lowered effectively the compution andcommunication cost of the global trust model.Thirdly, aiming at the deceitful trust evaluation and strategic oscillating behaviorof malice nodes, this thesis proposes two effective suppressing mechanismsrespectively which can enhance the anti-attack ability of the global trust model. Theevaluation management nodes are used to monitor the evaluation behavior of othernodes and compute their evaluation credibility, and the trust management nodes areused to monitor the oscillating behavior of other nodes and compute their oscillatingdegree. The model integrates evaluation credibility and behavior oscillating degreeinto local trust computation via a feedback control mechanism, and influence theglobal trust value of nodes indirectly. These nodes will be subjected to somepunishment of the system that their global trust value will be lowered whether theirevaluation credibility is low or their behavior oscillating degree is large, and vice versa. Simulation results show that the proposed model has the stronger anti-attackability and higher service success rate than the existing models related.Fourthly, this thesis proposes an incentive mechanism based on the global trustvalue of nodes to deal with the free-riding problem. After analysing the TfT(Tit-for-Tat) strategy of BitTorrent system, this thesis points out that the TfT strategyis practical in the ideal network where repeatable game can been established. In thelarge-scale P2P network with diverse services, however, the randomicity of tradebetween nodes results in the probability of building up a repeated game is very low.Therefore, the feasibility of the incentive mechanism based on direct reciprocity isquestionable. Since the global trust value of nodes can well reflect their contributionand service quality to networks, this thesis establishes service response, queryforward and neighbor selection rules according to the differences of global trust valuebetween nodes to encourage nodes to contribute actively and service dependablywhich reflects the TfT strategy indirectly. Simulations show that the incentivemechanism in this thesis can change the behavior strategy of the free-riders andencourage them to share resources and has good universality.Finally, a P2P file sharing prototype system integrating trust management andincentive mechanism is developed to verify the findings in this thesis.