Learning Reliability Models of Grid Resource Supplying

Resource exchange between Grid participants is at the core of Grid computing. Distributed bartering is a distributed and moneyless method of resource exchange. Recent work related to distributed bartering has mainly dealt with resource supplying. However, Grid participants still face an unstable resource environment due to the partial and intermittent nature of the exchanged resources. The problem considered in this paper is the unreliability of resource supplying. Though it cannot be totally avoided, a proactive stance may lower its impact in the long run. We propose to explore the reduction of performance variability by improving resource consumption. The goal is to enable Grid participants to identify and avoid unreliable resource suppliers by learning reliability models of resource supplying. A Machine Learning problem is defined and the generated models are applied to select more reliable resources in the hope of improving resource consumption.