Reputation-based Mechanisms for Evolutionary Master-Worker Computing

Abstract

We consider Internet-based Master-Worker task computing systems,such as SETI@home, where a master sends tasks to potentially unreliable workers, and the workers execute and report back the result. We model such computations using evolutionary dynamics and consider three type of workers: altruistic,malicious and rational. Altruistic workers always compute and return the correct result, malicious workers always return an incorrect result, and rational(selfish)workers decide to be truthful or to cheat, based on the strategy that increases their benefit. The goal of the master is to reach eventual correctness, that is, reach a state of the computation that always receives the correct results. To this respect, we propose a mechanism that uses reinforcement learning to induce a correct behavior to rational workers; to cope with malice we employ reputation schemes.We analyze our reputation-based mechanism modeling it as a Markov chain and we give provable guarantees under which truthful behavior can be ensured. Simulation results, obtained using parameter values that are likely to occur in practice, reveal interesting trade-offs between various metrics, parameters and reputation types, affecting cost, time of convergence to a truthful behavior and tolerance to cheaters.