摘要： 为改进同构应用在计算网格中的执行性能,提出了子作业指派方法。对于计算密集的应用,任务间通信是可忽略的,故一个这样的作业被划分为若干子作业,不同的子作业被分别指派到不同的机群,该作业划分是根据网格负载平衡完成的。非计算密集的应用在多站点计算时很少取得令人满意的性能,故一个这样的作业被整体指派到某个机群。为找出最适合机群,对每个机群的处理机性能和处理机间通信性能进行测量,并根据应用性能模型预测作业运行时间。实验显示,该子作业指派方法在优化同构应用的执行性能上是有效的。

关键词: 网格, 调度, 同构应用, 作业划分, 子作业指派

Abstract: To improve the execution performance of homogeneous applications in computational grid, sub-job assignment methods are proposed. For computation-intensive applications, inter-task communication is negligible, so such a job is divided into some sub-jobs, and different sub-jobs are assigned onto different clusters respectively. Here, job partition is made by grid load balancing. Non-computation-intensive applications seldom obtained satisfying performance in multi-site computing, so such a job is assigned onto a certain cluster as a whole. To find the most appropriate cluster, processor performance and inter-processor communication performance both are measured for every cluster, and job runtimes are evaluated according to application performance model. Experiments show that sub-job assignment methods are effective for optimizing execution performance of homogeneous applications.

中图分类号: 

• TP393