Uncheatable Reputation for Distributed Computation Markets

Reputation systems aggregate mutual feedback of interacting peers into a “reputation” metric for each participant. This is then available to prospective service “requesters” (clients) for the purpose of evaluation and subsequent selection of potential service “providers” (servers). For a reputation framework to be effective, it is paramount for both the individual feedback and the reputation storage mechanisms to be trusted and able to deal with faulty behavior of participants such as “ballot stuffing” (un-earned positive feedback) and “bad-mouthing” (incorrect negative feedback). While, in human-driven (e.g. Ebay) environments, these issues are dealt with by hired personnel, on a case by case basis, in automated environments, this ad-hoc manner of handling is likely not acceptable. Stronger, secure mechanisms of trust are required. In this paper we propose a solution for securing reputation mechanisms in computing markets and grids where servers offer and clients demand compute services. We introduce threshold witnessing, a mechanism in which a minimal set of “witnesses” provide service interaction feedback and sign associated ratings for the interacting parties. This endows traditional feedback rating with trust while handling both “ballot-stuffing” and “badmouthing” attacks. Witnessing relies on a challenge-response protocol in which servers provide verifiable computation execution proofs. An added benefit is an assurance of computation result correctness.