桌面网格中面向检查点的任务调度策略研究

摘要

In order to improve the resource usage rate in desktop grid, a shared-checkpoint management with several scheduling methods to decrease turnaround time of parallel applications, like FCFS, FCFS-AT (adaptive timeout), FCFS-TR (task replication) and FCFS-PD (prediction on demand) are presented and described. Several scheduling policies oriented toward fast turnaround times are also described and discussed by reducing negative effects of resources volatility on the executions of parallel applications. Through tracebased approaches over the resources of hundreds machines of an academic environment, a serial of experiments are done. Results show most scheduling policies performed better than the classical FCFS methodology. The effects of checkpoint frequency over turnaround time are minors.%为了提高桌面网格中资源的利用率,在FCFS(first come first service)调度的基础上,利用了检查点技术,提出与描述了几种高效的任务调度策略:FCFS-AT(adaptive timeout),FCFS-TR(task replication),FCFS-PD(prediction on demand)等.分析了调度策略对资源挥发性与不可用所带来的负面影响,建立了实验环境,采用对子任务数据跟踪的工具,在不同的设置条件下对这些调度策略上进行了测试.测试结果表明,大部分的检查点调度策略都比经典的FCFS算法性能好,检查点的数目也影响系统的性能,但是影响不是很明显.