| Summary: | 碩士 === 國立高雄應用科技大學 === 電機工程系碩士班 === 95 === Computational grids offer a powerful computational capability for data-intensive applications. However, grid resources usually are dynamic and non-dedicated to a specific user or application. These resource characteristics will cause that the jobs of resource owners and the jobs of grid users compete the same resources, and then the performance of all of the jobs will be degraded. For avoiding this problem, it is necessary to select resources for the execution of grid jobs. Although many past studies were devoted to attack the problem of resource selection, the proposed methods usually considered only the instant or short-term workload states of resources. However, most of grid applications usually need to be executed in a long time. Therefore, it is not good enough to consider only the short-term workload states into resource selection. For this reason, we propose a new resource selection method based on sequential workload patterns mining for grid applications. This method discovers the workload patterns of owners from the history of workload states of grid resources, and predicts the availability of each grid resource in a specific time period, and then selects a set of resources for executing grid jobs based on the predicted availabilities.
|
|---|
