HITGRID:基于动态规划技术的网格任务调度中间件

HITGRID:a Dynamic Programming based Scheduling Middleware in Grid Environment

随着计算规模不断扩大,人们对并行计算调度的性能要求也不断提高。网格技术的出现,解决了计算资源的共享性、异构性、规模可扩展性、鲁棒性以及安全性等方面的问题,但同时也在资源调度方面带来了新的挑战。HITGRID是一个基于Globus网格工具平台上的网格调度中间件,它接收用户提交的计算任务,针对用户对计算资源提出的要求,在网格环境中查找符合条件的资源进行调度。文中讨论了将NWS预测技术封装为标准网格资源信息的方法,以及资源调度的性能模型,给出了调度中资源选择的策略。最后通过一个计算大规模N体问题的实例表明笔者的工作是有成效的。

The larger the scale of computation is becoming,the higher performance of parallel computation is needed.The emerging Grid technology has developed a shareable,heterogeneous,scaleable,robust and secure way to use resource in a parallel environment.However,there are still some problems in resource scheduling.HITGRID is a scheduling middleware on Globus.It divides user's job into subjobs,and hands them out to proper resources in terms of the requirements of user.This paper discusses how to use NWS to forecast status as a Grid Information Provider,and the performance model of the scheduler.In final,it gives a comparison between this scheduler and others,using a large scale N-body problem as example.