DisTriB: Distributed Trust Management Model Based on Gossip Learning and Bayesian Networks in Collaborative Computing Systems

The interactions among peers in the Peer-to-Peer systems as a distributed collaborative systems are based on asynchronous and unreliable communications. Trust is an essential and facilitating component in these interactions specially in the uncertain environments. Various attacks are possible due to large-to-scale nature and openness of these systems that affects the trust. Peers has not enough information about other peers thus they can use trust management to reason about future interactions with other peers. In fact any peer needs to know the exact prediction of the trustworthiness of other peers. In the proposed approach the Bayesian network based trust management model is used to infer the trust value of task processor peers based on various aspects of their behaviors. The previous behavior of peers is considered to determine meticulous trust value of these peers in completely distributed environment. Since the trust is multifaceted concept, each aspect is evaluated using a single Bayesian network to provide finer-grained inference of trust. In the proposed model, DisTriB (Distributed Trust Management Model Based on Gossip Learning and Bayesian Networks), many aspects such as network link capacity and workload of task processor peers is used to construct the Bayesian network. The optimum time window size is found to obtain better performance too. Finally, a robust, asynchronous, gossip based protocol is proposed that can withstand high churn and failure rates, and can spread the trustworthiness of peers while the processing of tasks in the proposed collaborative computing system is performed. Simulation results shows that the proposed approach outperforms previous works.