【作者】 陈锦言；

【导师】 孙济洲；

【作者基本信息】 天津大学 ， 计算机应用技术， 2008， 博士

【摘要】 正确对P2P中的节点信誉进行评价,可以及时隔离恶意节点,保证其它节点的安全,提高P2P系统服务的质量和效率。针对节点信誉评价向量缺乏准确计算方法的问题,提出并证明了在均值邻域附近可以用正态分布对Beta分布进行近似计算的理论,并以此为基础从概率论的角度提出了节点信誉度向量的计算公式,该公式可以简化节点信誉向量的计算复杂度。针对中心式P2P信誉评价体系容易受到协同作弊干扰的问题,提出了根据节点在系统中的表现计算节点间相似度的计算公式,以此为基础实现了对节点的聚类,因为诚实节点行为具有一致性,所以诚实节点会聚类为最大的一个节点集合,作弊节点根据作弊方式不同会聚类为若干个小的节点集合,最后整个系统信誉评价结果将以最大节点集合的观点为主,因此消除了协同作弊节点对系统信誉评价结果的干扰。针对聚类速度较慢的问题提出了先分组计算,再进行信誉度合并的算法。实验表明上述算法对节点协同作弊有较好的抵御效果。以传统De Bruijn图为架构的P2P系统可以保证网络具有较小的半径,但是节点加入、离开算法过于复杂,容易出现节点负载不均衡的现象。为此本文提出了De Bruijn路由图和冗余De Bruijn路由图的架构,并在此基础上实现了节点的加入、离开、路由算法。理论分析和实验表明该路由架构既保留了传统De Bruijn图的优点,又简化了节点的加入、离开操作,同时使分配在每个节点上的负载较平均。以冗余De Bruijn路由图为基础,提出了分布式P2P架构下的信誉评价体系:每个节点保留多条冗余路由,进行信誉查询的时候优先选择可信任路由节点构建信誉查询路径,并使用查询反馈结果对路由记录进行更新,随着查询次数的增加,节点不断更新路由表中各备选路由的可信度,查询准确度也随之提高。实验表明该算法具有较好的可扩展性和查询准确度,适合大规模网络环境下使用,同时以此为基础可以实现资源的可靠查找。更多还原

【Abstract】 It is necessary to evaluate the peer’s reputation efficiently and correctly in order to isolate malicious peers, protect the safety of other peers and improve the service quality and efficiency.A new theory that the beta distribution can be calculated approximately by the normal distribution by the average value area is proofed, a new method which can calculated the reputation vector of peers precisely is proposed based on this theory. To eliminate the calculation error caused by malicious collectives, a reputation evaluation method based on aggregation is proposed. The similarity among peers is calculated according to the reputation their give to the common neighbors and the reputation evaluation value they received from the common neighbors and the reputation they give to each other. Usually all the honest peers can be aggregated to one large group because of their action similarity, while the malicious peers will be aggregated to several small groups because of their different cheating methods. The evaluation result will base on the largest group, by this mean the affection of malicious peers is minimized. To improving the calculation efficiency .All the peers is divided into several groups and calculated respectively, then the reputation is calculated by merge these calculation result. Experiments show that this method can isolate the malicious collectives efficiently.The P2P structure based on traditional de bruijn graph has both less edges and shorter network diameter, while its main shortcoming is that it’s too complex to process the insert and leaving of peers, and it also often lead to unevenly distribution of resource among peers. For this reason the routing de bruijn graph and redundant de bruijn routing graph and a new leaving and insert algorithm of peers were proposed. Theoretic analysis and experimental results show that the redundant de bruijn routing graph can simplify the insert and leave operations, and it also can distribute the load among peers evenly.A distributed reputation evaluation method was proposed based on redundant de bruijn routing graph: every peer has redundant routing peers and the peers with higher credit will be selected priorily. The credit of the routing peers is updated by the feedback of the reputation query , so the query accuracy will also improve with the processing of query. Experiments show that the algorithm is suitable for large-scale use of the network environment, and on the base of reliable routing table a robust resouce locating method can be constructed.更多还原