对可分割的计算密集型大型作业在并行且不间断运行情况下的完工时间与作业分割粒度之间的关系进行研究.首先分析了子作业之间无通信和有通信两种情况下可分割计算密集型大型作业的完工时间和分割粒度的关系,然后对可分割计算密集型大型作业在专用网格资源上的完工时间与分割粒度的关系进行仿真.仿真结果显示,大型作业的完工时间随着分割粒度的增大先减小后增大;当单个子作业的计算时间和通信时间之比增大时,作业的分割粒度可以更细,作业完工时间的最小值减小.因此完工时间最优的作业分割粒度不能过粗或过细.%There have been many divisible compute-intensive grand-challenge jobs running on volunteer grid.The relationship between makespans of such grid jobs and granularities of jobs' partitioning was studied.Firstly, the relationship between makespan and partitioning granularity of compute-intensive jobs with and without communications between sub-jobs were analyzed theoretically.Then, the relationship between makespan and partitioning granularity of a job with and without communications between sub-jobs running on dedicated grid resources in parallel mode were simulated.The simulation results show that grand-challenge job's makespan decreases at first and then increases when granularity increases.Granularity can be more grain, and the best makespan will decrease when the ratio of computation time to the communication time of a sub-job increases.To optimize job's makespan, the job's partitioning granularity should not be too coarse or too fine.
展开▼
