Sustaining Incentive in Grid Resource Allocation: A Reinforcement Learning Approach

Abstract

Encouraging resource sharing and cooperation among different parties is one of the central goals of grid computing. In real environments, however, selfish or malicious nodes can seriously degrade the sharing and cooperation performance of a grid. To solve this problem, we propose QIA, a novel Q-learning based resource Allocation mechanism that sustains Incentive for every participating node. Exploiting an economic model, QIA recognizes the importance of trust factor when allocating resources. Each provider considers a combined metric, which is composed of the bid price and the trust value, of a requester when allocating its resources. The incomplete information is a key issue for a provider in determining the relative weight of bid price and trust value. We propose a reinforcement Q- learning technique to resolve the issue, which is able to adapt the dynamics of grid environments. We implemented QIA in a real grid test-bed, CROWN grid. Comprehensive experiments have been conducted, which demonstrate the efficacy of QIA.