Improving user experience of submitting jobs to HPC resources

Abstract

The growth of e-science and grid computing has presented new opportunities to researchers for whom access to large computational and data-storage resources is essential, offering a choice of resources on which to conduct such work. The selection of an appropriate resource, or resources, can have a significant impact on the rate of progress of such work, but only if all relevant properties of a job's structure are taken into consideration; this is not the case with currently-used resource brokering systems. This paper proposes a method by which resource-brokering can be improved for real-world engineering jobs comprising many constituent tasks, and demonstrates through means of simulation the improvement that such a method offers, which in some cases reduces average execution times of individual tasks by over thirty percent.