【作者】 鲍翊平；

【导师】 张维明； 姚莉；

【作者基本信息】 国防科学技术大学 ， 管理科学与工程， 2011， 博士

【摘要】 当前,P2P系统的安全和可用性问题已经成为制约P2P技术发展和应用的最大障碍之一,面对大量存在的恶意和自私行为,P2P的信任风险正急剧增大。信誉系统提供了一种有效的机制,它能够抑制节点的恶意行为,并激励节点积极参与交易,从而为P2P系统构建一个运行平稳、安全高效的网络环境。为了让信誉系统适应P2P系统的演化,在复杂多变的动态环境下仍能保证其有效性,从而最终促进P2P系统的进化,本文以P2P文件共享系统为背景,在深入分析P2P系统演化性的基础上,结合当前信誉系统研究的理论成果和不足,我们提出紧密相关的两个研究问题:如何构建有效的信任评估模型,以及如何调节信任评估模型参数和结构以提高其适应性的问题。围绕解决上述问题,本文主要的研究工作和成果如下。(1)提出了一种演化的信誉系统并构建了其框架鉴于P2P系统与生态系统的相似性,本文借鉴网络社会学和生态学的相关理论,提出了P2P社会生态系统的概念,并构建了一种基于“大分布、小分布”的混合式架构的演化的信誉系统ERS。(2)提出一种基于种群进化的信任认知模型PATrustP2P系统中节点间的信任关系一定程度上反映了人类社会中的人际信任关系,而人际信任的构建与维护则是一个演化的认知过程。鉴于此,我们跳出传统信任评估模型的建模思路,从生态系统的种群进化中得到启发,结合社会认知论的相关理论和方法,借鉴人类良好的认知习惯和丰富的认知经验,提出了一种基于种群进化的信任认知模型PATrust。PATrust采用选择、交叉和变异三种进化策略实现了对信任的有效认知。模型还引入了一种基于“言行一致”的单次交易满意度计算模型和推荐信息协同过滤机制,前者一定程度上降低了单次交易满意度计算的主观性;后者较大改善了推荐信息过滤和推荐信任值评估的效率和准确性。仿真分析表明,与经典的信任评估模型EigenTrust和P2PRep相比,PATrust在收敛水平、整体适应度(整体交易成功率)方面均有较大提高,并能更有效地抑制更广泛类型的恶意节点,特别是具有适应能力的恶意节点的攻击行为。(3)提出一种基于元认知的PATrust参数适应性调节模型现有的一些信任评估模型一般采用在模型中引入确定性参数的方法实现模型在运行时的自适应调节,但是该方法存在参数设置的合理性有待验证、难以动态调节等方面的问题。针对该问题,我们借鉴认知调节和模糊逻辑控制原理,并利用人类在认知调节方面的丰富经验和良好习惯,为PATrust设计了一个基于元认知的参数适应性调节模型MARM。通过实时监测节点信任认知活动的进展,MARM能适时地对PATrust的参数进行动态调节。仿真分析表明,MARM调节下的PATrust在动态环境中表现出良好的性能。(4)提出一种基于集群的PATrust结构适应性调节模型鉴于现有基于分布式架构的信任评估模型在复杂性,通信、存储和计算代价,信任值收敛方面存在的问题,以及当前基于“大分布、小集中”的混合式架构的信任评估模型的不足,为改善PATrust的结构适应性,我们从生态系统的集群效应得到启发,提出一种新的基于集群的信任认知模型CBTrust。CBTrust通过基于集群的信任认知和向心力驱动的集群演化管理的相互作用,在获得良好激励效果的同时,也改进了信任认知模型的性能,降低了通信、存储和计算代价。仿真分析表明,相对于PATrust而言,CBTrust的性能得到了极大改善;通过集群演化管理,CBTrust还能有效激励节点积极参与交易。此外,抵御针对CBTrsut的攻击的能力的仿真也表明,CBTrust具有一定的健壮性。更多还原

【Abstract】 Currently,the security and availability problems of Peer-to-Peer(P2P) systems are becoming one of the biggest obstacles which Constraint the development and application of P2P technology, as there are lots of malicious and selfish behavior here, the trust risk for P2P is now rapidly increasing. In order to build a smooth running, secure and efficient network environment for P2P systems, the reputation system provides an effective mechanism, which can inhibit the peers’ malicious behavior, and incent them to actively participate into the transaction.In order to make the reputation system to adapt to the evolution of P2P systems, and can still ensure and improve its effectiveness in complex dynamic environment, and ultimately promotes the anagenesis of P2P systems. In view of this, under the background of P2P file sharing system, and on the basis of in-depth analyzing of the evolution of P2P systems, and with the current theoretical achievements and shortcomings, two closely related research questions are proposed in our thesis: how to build a more natural and more reasonable trust evaluating model; and how to adaptively regulate the parameter and structure of trust evaluating model. To solve those questions above, our work in this paper can be briefly introduced as following.(1)An Evolutive Reputation System and its FrameworkIn view of the similarity between P2P systems and ecosystem, and by using the related theories of Network Society and Ecology for reference, the concept of P2P social ecosystem (P2PSE) is proposed, where this paper presents an evolutive reputation system (ERS) based on hybrid architecture with the characteristics of "distributed in the whole, whereas distributed in the cluster".(2) A Population Anagenesis based Trust Cognition ModelAs the trust relationship between peers in P2P system to some extent reflects the interpersonal trust relationship in human society, and the build and maintenance of interpersonal trust is an evolutive cognitive process. In view of this, through thinking out of the traditional modeling for trust evaluating, and being inspired by the population anagenesis of ecological systems, and being associated with social cognitive theory and related method, a population anagenesis based trust cognition model (PATrust) is proposed. The main contributions of the PATrust include: good cognitive habits and rich cognitive experience of human beings are effectively utilized; three evolutionary strategies, i.e. selection, crossover and mutation are introduced so as to effectively cognize the peer’s trust; in order to reduce the subjectivity of the calculating of transaction satisfaction degree (TSD), a model for calculating TSD based on "practise what you preach" is proposed; through introducing the mechanism of collaborative filtering for recommendations, the Efficiency and accuracy of recommendation filtering and the recommending trust value evaluating are greatly improved.Our experiments show that the PATrust outperforms the traditional trust evaluating model, such as the EigenTrust and P2PRep, in terms of both Convergence and total fitness degree, and that our model can more effectively inhibit a broader type of malicious behavior, especially the attack behavior performed by some adaptable malicious peers.(3)A Metacognition-based Adaptive Regulating Model for the parameter of PATrustIn order to improve the adaptability of trust evaluating model, some existing models commonly utilize approaches by introducing the deterministic parameters, however, those approaches still have problems, such as: the rationality of parameter setting needs to validate; it’s hard to dynamically adjust the parameters at runtime. As a result, from the perspective of cognitive regulation and fuzzy logic control, and through using the good habits and rich experience of human beings in cognitive regulation, a metacognition-based adaptive regulating model for the parameter of PATrust (MARM) is proposed.Our experiments show that the PATrust with MARM outperforms the PATrust without MARM in dynamic environment.(4) A Cluster-based Adaptive Regulating Model for the structure of PATrust Most existing distributed architecture-based trust evaluating models have limitation in the complexity, the cost of communication, storage and computation, and the convergence of trust value. Furthermore, the trust evaluating models based on hybrid architecture with characteristics of "distributed in the whole, whereas centralized in the cluster" also have their limitation. We propose a cluster-based trust cognitive model (CBTrust). Through the interaction between the cluster-based trust cognition and centripetal force-based Cluster Evolution Management, the main contributions of the CBTrust include: the performance of trust cognitive model is improved; good incentive effect for peers is also obtained.Our experiments show that the performance of the CBTrust is greatly improved while comparing with the PATrust, and the CBTrust can effectively incent the peers to actively participate into the transaction, and the CBTrust is robust to resist the attack aimed at the CBTrust itself.更多还原