Research On Trust Mechanism And Overlay Construction In P2P Networks

Peer-to-Peer computing is an Internet computing paradigm which has been extensively studied in recent years. Its final goal is to utilize all kinds of possible end-devices in the Internet to construct cooperating and resource sharing application environments. In real large scale P2P overlays, the overlay dynamic and the peer heterogeneity in providing computing, storage and bandwidth services are the main challenges. Past studies and practices have shown that the main reason behind these challenges is because of the autonomousness of peers. While the peer autonomousness is represented by their behavior patterns in network, it is reasonable for P2P systems to handle dynamics and peer heterogeneity by their innate behavior patterns.Trust can be regarded as the peer behavior patterns in providing services. Therefore a proper definition of trust could be a reasonable metric to differentiating the behavior patterns of peers. In P2P environments, a common way for trust generation and management is by constructing trust (or reputation) systems, which involves trust modeling and trust management.Based on above considerations on how P2P systems handle network dynamics and peer heterogeneity by trust, we start our studies from two main kinds of trust models, namely the probabilistic trust model and the social network based trust model. In the probabilistic trust models, we focus on effective trust modeling without using global information about peer's behavior patterns. In social network based trust models, we target a more general trust model by introducing peer similarities in providing feedbacks on trust evaluations. Finally, we use the developed probabilistic trust mechanism in P2P overlay construction that effectively employs the peers trust heterogeneity and as a consequence results in a boost in the overall overlay performance and efficiency. The major contributions of this work are as follows:1. Probabilistic trust modeling: we present a kalman-filter based trust model named RealRep. RealRep is a recommendation based probabilistic trust model which implements the separation between direct services and recommendation services through different trust modeling and evaluating strategies. The differences between RealRep and other probabilistic trust model are that RealRep relaxes the limitation on assumptions of peer behavior patterns, thus allowing trust evaluation on more complicated behaviors. Additionally, RealRep avoids the dependence on the global information, such as the rate of untrustworthiness peers in the networks.2. Peer similarity based trust modeling: we incorporate peer similarity measurement and feedback weighting techniques in social network based trust model. For tackling the data sparseness problem in P2P environments when measuring peer similarity, we introduce the similarity prorogation algorithm, named Maximin, to allow similarity construction between two nodes without direct interactions by exploring all the similar peers between them. And the simulation results show the effectiveness of the proposed algorithm. We further use similarity measurement in global trust generation by weighting the feedbacks according to the similarities between the feedback provider and the trust evaluating peer.3. Hierarchical overlay construction: we introduce a trust enhanced unstructured overlay adaptation protocol, through which peers in the network select the best strategy on selecting neighbors by exploring the peers'trust heterogeneity. This results in an overall performance increase. We also argue that trust can be a metric in selecting superpeers for hierarchical overlay construction. In hierarchical overlay construction, one of the main problems is the extra overload introduced in superpeer selection and above overlay construction. For this, we present a new superpeer selection and competition protocol for effective overlay construction. The protocol results in an overlay construction by fewer overloads in messages and better performance in superpeer overload balance.