Using Frequent Workload Patterns in Resource Selection for Grid Jobs

Resource selection is an important issue of grid computing. If a grid job can stably gain enough CPU time from the same resources, not only the execution time of the job but also the frequency of resource reallocation is effectively minimized. However, most of the proposed methods are not effective enough to resolve the problem of resource selection in computational grids. The main reason is that these methods usually make use of current workload state or short-term prediction in available CPU time to be the basis of resource selection while most of grid jobs require a long execution time. To address this problem, we propose a novel algorithm of resource selection for computational grids in this paper. The basic concept of this algorithm is to discover the frequent workload patterns of resources, and then select resources for grid jobs according to the long-term prediction of resource availability by using frequent workload patterns.