Download files
Abstract
The Reputation-Based Trust mechanism helps a service assess the trustworthiness of offered services, based on the feedback obtained from their users. However, in the absence of appropriate safeguards, service users can manipulate this feedback, and existing attempts to encourage truthful feedback fail when the majority of users are potentially dishonest. Existing approaches also suffer from a lack of suitable distributed architectures to support their integration within trust systems, so that their overhead costs may be excessive. This thesis proposes novel distributed architectures for interfacing an auction mechanism (one that has already addressed the problem of manipulation by market-trading participants) with trust systems. These architectures will discourage users from giving untruthful feedback and minimize overhead costs in integrating the mechanism into trust systems.
This research follows an empirical research methodology that uses case studies inspired by real-world scenarios in Service-Oriented Computing (SOC). The research solution has two aspects: it proposes a technique for mapping the concepts of auction theories from economics into trust-based domains; and it draws on the concepts of an architecture-based approach to facilitate the integration of auction components within existing trust components. This facilitation aims to achieve architectural properties for interfacing: extensibility, to keep trust systems’ normal functionalities when they are enhanced with the auction mechanism's capabilities; scalability, to ensure highly scalable trust systems; decentralization, to capture an auction's truth-telling property in fully distributed environments such as SOC; and performance, to ensure an affordable response time while achieving scalability. The research solution is evaluated through three case studies, namely centralized, decentralized and hybrid computing. Each of these case studies is progressively built up from the architecture of the previous case, thus permitting comparative analysis of: (1) the feasibility of utilizing the mechanism in trust-based domains; and (2) the architecture supporting the mechanism based on architectural properties for interfacing. The analysis is based on metrics including users’ utility gains, CPU and memory usage, and a system's response time.
The contributions to research are the proposed mapping technique and the three architectures (one for each case study) supporting the integrated auction mechanism. A trust-based engine with an integrated auction mechanism is implemented to support the approach. The case-study results demonstrate that the architectures can ensure truthful reporting by users even when up to 90% of them are dishonest. In centralized computing, it can achieve architectural solutions regarding extensibility with less significant overhead (increases of 5% in CPU usage, 1.93% in memory usage, and 6% in response time); in decentralized computing, it can achieve both scalability (three times higher performance compared to centralized computing in equivalent environments) and decentralization (the auction's truth-telling property captured); and in hybrid computing, it can achieve trade-offs between scalability and performance (response time reduced by one-third compared to the scalable equivalent in decentralized computing).
The three architectures with the concepts of the proposed mapping technique can ensure service users have an incentive to be truthful even when untruthful users are in the majority, while also achieving architectural properties for interfacing. Each of these architectures is suitable for a certain situation: the centralized one for small-scale service systems, the decentralized one for highly scalable service systems, and the hybrid one for service systems that have to balance trade-offs between scalability and performance.