Distributed Trust Management in Grid Computing Environments

Grid computing environments are open distributed systems in which autonomous participants collaborate with each other using specific mechanisms and protocols. In general, the participants have different aims and objectives, can join and leave the Grid environment any time, have different capabilities for offering services, and often do not have sufficient knowledge about their collaboration partners. As a result, it is quite difficult to rely on the outcome of the collaboration process. Furthermore, the overall decision whether to rely at all on a collaboration partner or not may be affected by other non-functional aspects that cannot be generally determined for every possible situation, but should rather be under the control of the user when requesting such a decision. In this thesis, the idea that trust is the major requirement for enabling collaboration among partners in Grid environments is investigated. The probability for a successful future interaction among partners is considered as closely related to the mutual trust values the partners assign to each other. Thus, the level of trust represents the level of intention of Grid participants to collaborate. Trust is classified into two categories: identity trust and behavior trust. Identity trust is concerned with verifying the authenticity of an interaction partner, whereas behavior trust deals with the trustworthiness of an interaction partner. In order to calculate the identity trust, a "small-worlds"-like scheme is proposed. The overall behavior trust of an interaction partner is built up by considering several factors, such as accuracy or reliability. These factors of behavior trust are continuously tested and verified. In this way, a history of past collaborations that is used for future decisions on further collaborations between collaboration partners is collected. This kind of experience is also shared as recommendations to other participants. An interesting problem analysed is the difficulty of discovering the "real" behavior of an interaction partner from the "observed" behavior. If there are behavioral deviations, then it is not clear under what circumstances the deviating behavior of a partner is going to be tolerated. Issues involved in managing behavior trust of Grid participants are investigated and an approach based on the idea of using statistical methods of quality assurance for identifying the "real" behavior of a participant during an interaction and for "keeping" the behavior of the participants "in-control" is proposed. Another problem addressed is the security in Grid environments. Grids are designed to provide access and control over enormous remote computational resources, storage devices and scientific instruments. The information exchanged, saved or processed can be quite valuable and thus, a Grid is an attractive target for attacks to extract this information. Here, the confidentiality of the communication between Grid participants, together with issues related to authorization, integrity, management and non-repudiation are considered. A hybrid message level encryption scheme for securing the communication between Grid participants is proposed. It is based on a combination of two asymmetric cryptographic techniques, a variant of Public Key Infrastructure (PKI) and Certificateless Public Key Cryptography (CL-PKC). The different methods to trust management are implemented on a simulation infrastructure. The proposed system architecture can be configured to the domain specific trust requirements by the use of several separate trust profiles covering the entire lifecycle of trust establishment and management. Different experiments illustrate further how Grid participants can build, manage and evolve trust between them in order to have a successful collaboration. Although the approach is basically conceived for Grid environments, it is generic enough to be used for establishing and managing trust in many Grid-like distributed environments