网格工作流中基于优先级的调度方法研究

Study on Scheduling Methods Based on Priorities in Grid Workflow

根据网格用户的身份、信誉,以及网格工作流中应用程序的时间紧迫度和依赖关系,提出了网格工作流中应用程序的优先级计算方法.其中,完全优先级调度算法根据网格工作流中的应用程序优先级向量生成调度序列,而且每次只将队列中优先级最高的应用程序提交给网格,而非完全优先级调度算法可同时调度若干无依赖的网格应用程序进入网格,以弥补完全优先级调度算法的并行化问题.实验表明,当网格中的资源较少且资源的性能差异较大时,使用完全优先级调度算法可以保证优先级较高的工作流的应用程序优先使用优势资源;当网格中的资源性能差异不大时,非完全优先级调度算法可解决因资源未充分利用而造成工作流完成时间大幅增加的问题.

A priority calculating method was proposed, in which the grid workflow applications＇ priority vector was calculated by the users＇ identity, credit, urgency and dependency of grid applications in the workflow. The full priority scheduling algorithm generates the scheduling list according to the priority vectors, and only the application which has the highest priority can be submitted while the non-absolute priority scheduling algorithm schedules several independent grid applications so as to compensate the parallelism problem of the full priority scheduling algorithm. Experiments show that, when the grid has fewer available resources or the performance discrepancy of the resources is much higher, the full priority scheduling algorithm ensures the workflow applications that have high priority to be scheduled firstly, and when the grid contains more resources that have the same performance, the non-absolute priority scheduling algorithm avoids largely increasing of the workflow＇s makespan caused by resources waste.