A framework for decentralised trust reasoning

• Main Author: Abdul-Rahman, Alfare
• Published: University College London (University of London) 2005
• Subjects: 005.8
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.420984

Recent developments in the pervasiveness and mobility of computer systems in open computer networks have invalidated traditional assumptions about trust in computer communications security. In a fundamentally decentralised and open network such as the Internet, the responsibility for answering the question of whether one can trust another entity on the network now lies with the individual agent, and not a priori a decision to be governed by a central authority. Online agents represent users' digital identities. Thus, we believe that it is reasonable to explore social models of trust for secure agent communication. The thesis of this work is that it is feasible to design and formalise a dynamic model of trust for secure communications based on the properties of social trust. In showing this, we divide this work into two phases. The aim of the first is to understand the properties and dynamics of social trust and its role in computer systems. To this end, a thorough review of trust, and its supporting concept, reputation, in the social sciences was carried out. We followed this by a rigorous analysis of current trust models, comparing their properties with those of social trust. We found that current models were designed in an ad-hoc basis, with regards to trust properties. The aim of the second phase is to build a framework for trust reasoning in distributed systems. Knowledge from the previous phase is used to design and formally specify, in Z, a computational trust model. A simple model for the communication of recommendations, the recommendation protocol, is also outlined to complement the model. Finally an analysis of possible threats to the model is carried out. Elements of this work have been incorporated into Sun's JXTA framework and Ericsson Research's prototype trust model.