Logo image
探究網格計算中最後完工期限對於排程的必要性
Thesis

探究網格計算中最後完工期限對於排程的必要性

曾莉雅
Masters, National Tsing Hua University
2009

Abstract

排程啟發式法總加權延遲時間指派時間最後完工期限總完工時間網格計算 Scheduling heuristicTotal weighted tardinessAssignment timeDeadlineMakespanGrid computing
Many scheduling heuristics have been dedicated to either Heterogeneous Computing (HC) or Grid computing for the purpose of minimizing makespan, instead of the reduction of cost or deadline of a job. In the last few years, several researches have devoted to the study of scheduling heuristics within deadline factor. However, they treated deadline as a constraint that a task is completed by its deadline, and the maximum total number of completed tasks meeting deadlines is the main objective of these researches. Otherwise, some researches dealt deadline with a weight that the maximum evaluation value is the motive of them. Deadline of a task usually be utilized by other field sciences, such as Project Management and Operations Management, this study considers whether the utilization of deadline affects makespan in Grid computing systems is advantageous should be confirmed. Additional, both the number and the heterogeneities of computers found in Grid computing are more complex than in HC. The relevant HC researches are based on high-speed bandwidth and the assignment time of delivering data has usually been ignored, so the relevant heuristics of HC may not proper for Grid computing reasonably.Therefore, in response to the appeal to two conditions: 1) minimum makespan under reduce cost; and 2) minimum makespan under deadline. This study simulates two parts: 1) the Apparent Tardiness Cost Setups-Minimum Completion Time (ATCS-MCT) scheduling heuristic considers both makespan and cost; and 2) utilize task deadline and task assignment time to testify that makespans of the proposed scheduling heuristics are better than original scheduling heuristics. Furthermore, it can be verified that utilizing the deadline of each task is really advantageous in a dynamic Grid computing environment.

Metrics

1 Record Views

Details

Logo image