SORT: A SELF-ORGANIZING TRUST MODEL FOR PEER-TO-PEER SYSTEMS

This paper presents distributed algorithms used by a peer to reason about trustworthiness of other peers based on the available local information which includes past interactions and recommendations received from others. Peers collaborate to establish trust among each other without using a priori information or a trusted third party. A peer’s trustworthiness in providing services, e.g., uploading files, and giving recommendations is evaluated in service and recommendation contexts. Three main trust metrics, reputation, service trust, and recommendation trust, are defined to precisely measure trustworthiness in these contexts. An interaction is evaluated based on three parameters: satisfaction, weight, and fading effect. When evaluating a recommendation, including to these parameters, recommender’s trustworthiness and confidence about the information provided are considered. A file sharing application is simulated to understand capabilities of the proposed algorithms in mitigating attacks. For realism, peer and resource parameters are based on several empirical studies. Service and recommendation based attacks are simulated. Nine different behavior models representing individual, collaborative, and identity changing malicious peers are studied in the experiments. Observations demonstrate that malicious peers are identified by good peers. The attacks are mitigated even if they gain high reputation. Collaborative recommendation-based attacks might be successful when malicious peers make discrimination among good peers. Identity changing is not a good attack strategy.