Research On Efficiency And Security In Unstructured Peer-to-Peer Networks

In recent years, peer-to-peer (P2P) networks have become the applications which dominate the traffic bandwidth of Internet. Comparing with the previously proposed Client/Server (C/S) model, P2P network is self-organized and distributed, which can make full use of the capability of each peer to publish or retrieve information on the purpose of sharing resources effectively and improving resource utilization. However, P2P network runs in a highly dynamic environment where peers can join or leave P2P network at any moment, and the popularity of the sharing resources varies with time. Peer's heterogeneity further raises the level of complexity in providing efficient access to existing resources. These facts significantly affect the efficiency of resource sharing. Besides, due to the anonymous and autonomous of peers in decentralized P2P network, they may involve in some security problems when they have interactions. The efficiency and security are the two main performances, which directly decide the practicability of P2P network.The blind flooding algorithm under a fixed time-to-live (TTL) constraint using in unstructured peer-to-peer networks, may cause a large amount of network traffic, as well as no guarantee for the hit rate of resources which exist in the network, especially for rare and distant ones. To mitigate these problems, PercolationNET, a multi-tree sub-overlay, which is built on top of the degree distribution of existing P2P overlay (named original overlay) is proposed. PercolationNET organizes peers in a tree-based structure which is reliable and efficient for message dissemination. Each tree comprises peers and links existing in the original overlay, and each tree root is a super peer which owns high degree in the original overlay. Correspondingly, the search process can be divided into two stages. In the first stage, a query message is flooded with an appropriate TTL value in the original overlay network so that the message can spread to all trees of the sub-overlay with smaller redundant messages. Then the query message is broadcast by flooding along the sub-overlay in the second stage, which has low connectivity but ensures that any object existing in the network can be found.One of the essential problems in unstructured peer-to-peer network is the strategy of resource discovery. Previous researches either have poor response speed, or generate too much network traffic. In order to reduce the traffic load and improve the availability of the sharing resources in unstructured P2P networks, a caching scheme combining alternative index and adaptive replication (AIAR) is presented. AIAR uses random walk mechanism to disperse caching information of resources in the network based on its Power-law characteristic, and dynamically adjusts replicas according to the visit frequency on resources and the degree information of peers.In P2P file sharing systems, the participating peers sharing the files with others. However, due to anonymity and openness, there are many trust and free-riding problems such as malicious peers and free-riders in P2P file sharing systems. A remedy for these problems by introducing access control to trust mechanism for P2P file sharing systems is proposed. First, the thesis defines a general trust metric which integrates the following parameters: direct trust between the requestor and the provider, recommendation from other peers and time declining factor. Moreover, each shared file has two threshold values to control access. One is the credibility threshold value and the other is the contribution threshold value. The credibility threshold value is used to guarantee that only when the credibility of a peer is bigger than a certain value, the peer can download a file from the provider. The contribution threshold value is used to restrict that only the peers who have ever provided enough download service to a peer can download files from it. For each requesting peer, it can not access a file until it satisfies accessing qualification of the file. Especially, this model can isolate malicious peer effectively.How to minimize threats in an open community is an important research topic. Trust models have been widely used in estimating the trustworthiness of peers in P2P systems where peers can transact with each other without prior experience. However, current P2P trust models almost take no consideration for the nature of trust—fuzzy, complex and dynamic, which results in low efficiency in resisting the attacks of malicious nodes. a new trust model—NatureTrust, which can alleviate the shortage brought by the nature of trust, is designed. In order to cope with the fuzzy characteristic of trust, linguistic terms are used to express trust. Additionally, fuzzy inference rules are employed to evaluate trust of each transaction so as to handle the complex characteristic of trust. Furthermore, risk factor is deployed into NatureTrust to represent and reason with the dynamic characteristic of trust. Both risk and trust factors are considered in evaluating the trustworthiness of each peer.