Research On Reputation Based Trust Management In Internet-based Virtual Computing Enviroment

Internet-based virtual computing environment (iVCE) is a new research approach to constructing a "trustworthy computing platform" on Internet. iVCE aims to systematically provide harmoniously unified, secure, and transparent services to Internet applications by aggregating and integrally utilizing of widespread resources. Accommodating to the growing and autonomic nature of Internet resources, it has been recognized as an important way to aggregating and utilizing Internet resource that building self-organized iVCE by voluntary participating and autonomic coordinating of the user nodes. The practice of the typical self-organizing system, such as P2P system, demonstrates the great potential of this way. The open, dynamic and autonomous nature of the self-organizing systems brings up serious behavior trustworthiness problems, such as unreliable and fraud services, free-riding and resource abusing. Reputation based trust management provides an important way to solve the behavior trustworthiness problems in self-organized iVCE environment. It can help peers to evaluate trustworthiness of each other, thus improve the quality of interaction, refrain the impact of unreliable and fraud service and guarantee the work of the system healthily, meanwhile, it can encourage rational peers to provide good quality service, pernalize malicious peers and improve the availability of the system.Reputation based trust management is still a developing technology and there are many problems need to be studied, such as weakness in modeling dynamic trust relationship and aggregating trust information, lack of effective mechanism promoting peers give honest recommendations actively, little effort to study reputation based incentive mechanism in the certain network application environment. To solve these problems, the dessertation firstly analyzes the research work on reputation based trust management comprehensively, then focuses the problems of valuation and elicitation of peers' behavior in self-organizing system and deeply studies trust model, honest recommendation incentive mechanism and reputation based incentive mechanism. Main works of the dissertation are as follows:(1) To overcome the weakness of the existing trust model in countering strategically altering behavior and aggregating feedback information, we propose a time-frame based dynamic trust model DyTrust. We incorporate time dimension with time-frame, which captures experience and feedback's time sensitivity. We also introduce four trust parameters in computing trustworthiness of peers, namely, short time trust, long time trust, misusing trust accumulation and feedback credibility. Together, these parameters are adjusted in time to reflect the dynamics of the trust environment using feedback control mechanism, thus, the trust evaluation has better adaptability to quick changes in peers' behavior. On the strategies used for implementing the trust model, we use P-Grid to manage the trust information efficiently, and use the proof of interaction, PKI based scheme and data replication to increase the security and reliability of trust model implementation. Theoretical analysis and simulation show that, DyTrust has advantages in modeling dynamic trust relationship and aggregating feedback information over the existing trust metrics. It is highly effective in countering malicious peers regarding strategically altering behavior and dishonest feedbacks of malicious peers.(2) Current trust model based on localized reputation information can not process such strategic recommendations as correlative and collusive ratings. Furthermore there exists unfairness to blameless peers in these models. Then, this thesis presents a robust trust model ARTrust used by a peer to reason about trustworthiness of other peers based on the available local information which includes past interactions and recommendations received from others. It proposes a recommendation construction approach in which recommendation consists of the direct trust value and level of confidence about this value, then a recommendation evaluation scheme is defined. When evaluating a recommendation, recommender's recommendation credibility and confidence about the information provided are considered. Theoretical analysis and simulation show that ARTrust can help peers effectively detect dishonest recommendations in a variety of scenarios where more complex malicious strategies are introduced, and achieve a more accurate trust valuation and fair evaluation of recommendations.(3) In order to stimulate reputation information sharing and honest recommendation elicitation, this thesis presents an incentive mechanism. Based on an extension of ARTrust model, incentives for participation and honest recommendation is implemented through a fair differential service mechanism. To achieve this goal, first, we define two parameters that can be used to create service differentiation in trust information exchange, namely, level of participation and recommendation credibility. Second, based on the above two parameters, we propose a simple yet effective reputation information exchange protocol to elicit sufficient and honest participation. Theoretical analysis and simulation show that, as a supplementary component of ARTrust, it can stimulate peers to send sufficiently honest recommendations.(4) In self-organizing system, the service availability is seriously affected by misbehavior of selfish and malicious peers. So, we propose a reputation based incentive mechanism for unstructured P2P systems. We can valuate peer's service reliability using the trust model ARTrust mentioned above. In order to reward the good behavior and penalize the bad by using trust value effectively, first, we propose an effective topology construction algorithm. Second, we select proper policies that determine the pairs of peers eligible to interact with each other, namely "provider selection" and "contention resolution". Theoretical analysis and simulation show that the reputation based incentive mechanism is simple yet effective in combating the misbehavior of selfish and malicious peers. Moreover, it has favorable feasibility of implementation.