Research On Trust Management In Collaborative Computing Systems

Trust and reputation management still plays an active role in collaborative computing systems. A common challenge confronting collaborative computing system is how to effectively mitigate the attacks through trust management. Some participants (peers) in most collaborative systems may have little knowledge about other participants with whom there is no prior interaction or transactional experiences. Thus, reputation-based trust management has been utilized as an effective service selection criterion to evaluate how much one can trust others. The existing of malicious participants makes it hard for trust management to evaluate participants’transactional behaviors accurately and rationally. Malicious participants can manipulate the system by getting high reputation through cheat, collusion, disguise, Sybil attack and spy etc. Considering the malicious behaviors of strategic participants, we propose related theories and metrics to address in terms of peer roles, peer cluster and trust propagation. Our main contributions are as follows:(1) In collaborative computing systems, each peer can have two kinds of behavior roles: service producer (provider) and service consumer. Service producer provides service for consumer and receives trust (feedback) ratings; the service consumer rates the transaction quality after accomplishing the interaction. Based on these two kinds of behavior roles played by each peer, this paper proposes a peer-behavior based trust mechanism Dual-EigenRep. Our mechanism establishes two corresponding reputation values for these two kinds of behavior roles, and makes these two reputation values interact and reinforce mutually. The good peers will form a high-reputation community, in this way those malicious peers are isolated effectively.(2) Collaborative computing systems still face some fundamental challenges, for instance, how to identify and isolate the malicious peers? How to resist the strategic malicious peers? What trust metrics are implemented to guarantee the service providers are authentic? Keeping these questions in our mind, we propose a maximum flow theory based Peer-Cluster mechanism. This mechanism utilizes the maximum flow theory to make authentic peers cluster together. Peers in this cluster would have priorities to provide services for other requesters, which can restrain the transactional behaviors of malicious peers validly even they have high reputation values. We prove the existance of this cluster in theory and the correctness of our algorithm through experiments. (3) In terms of the vulnerabilities that the uniform trust propagation boosts the reputation values of malicious peers, and degrades the reputation values of good peers in EigenTrust. We do a deep investigation and identify the inherent vulnerabilities and root causes on local trust computation, trust-feedback credibility, trust propagation scheme and trust-feedback network density. To address these vulnerabilities, we propose attack resilient trust propagation metrics and similarity based feedback credibility and weighted local trust. Our metrics utilize uniform and conditioned trust propagation manners to compute the global trust. Extensive experiments also show that our metrics can effectively constrain the malicious peers to gain high trust propagation from good peers.