群机管理系统的作业调度算法研究

Study on Scheduling Policy of Cluster Computing Management

首先分析了常用的作业调度策略及单独使用各策略的不足，然后介绍了作者在自行开发设计的异构群机计算管理系统TUMBLER中提出和使用的三级作业调度策略，重点介绍了在确定作业优先级时使用的基于资源需求度概念的短需求优先的策略。经测试表明，采用基于短需求优先的三级调度策略时，TUMBLER系统较未采用短需求优先策略时的性能有明显提高。

With the development of networking technologies, more and more attentions are paid on the cluster computing management software, which is as the middleware of resources management. It is the key element for cluster system of high performance to schedule the jobs fairly and impartially according to the characteristics of cluster resources. This paper first introduces scheduling policies in common use, analyses the deficiencies of each policy, then describes the policies in TUMBLER system, which is on the basis of Shortest Demand First (SDF).