On the Optimization of GLite-Based Job Submission

A Grid is a very dynamic, complex and heterogeneous system, whose reliability can be adversely conditioned by several different factors such as communications and hardware faults, middleware bugs or wrong configurations due to human errors. As the infrastructure scales, spanning a large number of sites, each hosting hundreds or thousands of hosts/resources, the occurrence of runtime faults following job submission becomes a very frequent and phenomenon. Therefore, fault avoidance becomes a fundamental aim in modern Grids since the dependability of individual resources spread upon widely distributed computing infrastructures and often used outside of their native organizational boundaries, cannot be guaranteed in any systematic way. Accordingly, we propose a simple job optimization solution based on a user-driven fault avoidance strategy. Such strategy starts from the introduction within the grid information system of several on-line service-monitoring metrics that can be used as specific hints to the workload management system for driving resource discovery operations according to a fault-free resource-scheduling plan. This solution, whose main goal is to minimize the execution time by avoiding execution failures, demonstrated to be very effective in incrementing both the user perceivable quality and the overall grid performance.