Class-Based Grid Resource Management Strategies for On-Demand Jobs

Grid computing has emerged as a new paradigm for distributed systems, which promotes sharing of distributed resources. To maximize its benefits, it is essential to discover the resources available on the grid, and then effectively map the jobs to the resources for maximizing a given objective function. This paper focuses on the problem of matching of jobs to resources in a computing grid. Jobs are classified based on their service demands. Matching policies that use only the knowledge of job classes are introduced in this paper; simulation experiments demonstrate the effectiveness of these policies. Under a variety of different workload parameters the proposed matching policies demonstrate a performance comparable to, or better than, the well-known Minimum Completion Time matching policy, which is based on detailed a priori knowledge of jobs and resource characteristics.