一种基于流作业的网格作业调度研究

A Grid Scheduling Approach Based on Stream Jobs

在基于网格环境的一些网格应用中,用户需要提交一种作业类型,该作业可以被分解为逻辑上独立的元作业,这些元作业不存在依赖和通讯关系,并且它们的执行需要大量的数据移动。针对这种作业类型,本文提出了一种基于流作业的网格调度模型。在该模型中,这些独立的元作业像"流"一样自主地流向各个计算节点去执行,各计算节点接收的流量取决于其计算能力,并避免"断流"问题。同时,该模型还分离了作业流和数据流,实现了作业逻辑控制和数据控制的分离,提高了调度的灵活性。本文将该调度模型应用于药物虚拟筛选应用中,该模型能够充分利用计算节点的计算能力。

In some grid applications, users need to submit a type of jobs which can be divided into independent metajobs in logical. There is no data dependence and communication between metajobs and there are a lot of data movements in their executing processes. For this type of jobs, we propose a grid scheduling model based on stream jobs. In this model, metajobs move to different computing nodes autonomously like a stream. The job flow depends on the computing capability of computing nodes and there is no “discontinuous flow” problem. Besides, in this mode the job stream and the data stream are separated, which realizes that the logical control of the job and the control of data are separated and the flexibility is improved. In this paper, the model is implemented in the virtual screening of drugs and it can make use of the nodes’ computing capability fully.