Research And Optimization Of Job Scheduling In Lsf System

Grid computing is a new development trend of distributed computing in fields of scientific research and commercial business. As a key component of grid computing, job management system provides efficient job scheduling strategies. It integrates and makes full use of distributed computing resources to improve the system utilization.In this paper, job management system LSF of Platform Computing Company is studied. At first, the architecture, daemons and process of job scheduling of LSF are introduced. Secondly, the job scheduling strategies of LSF are analyzed. They can not meet the demand of guaranteeing fairness of resource sharing among users. In order to solve this problem, the dynamic priority is introduced and the formula of user priority computing is proposed. The user-based fair scheduling strategy is designed and implemented based on the user dynamic priority in LSF. Thirdly, LSF has poor performance while it schedules large number of short jobs. Its system bottleneck is analyzed. In order to solve this problem, a new child scheduler is designed with the hierarchical architecture. For optimizing the LSF system architecture, the child scheduler is used to dispatch and manage jobs in parallel, so the system throughput and performance are enhanced. At last, the fair scheduling strategy and optimized LSF have been tested. It turned out that users can use the resources fairly with fair scheduling strategy and the system performance is enhanced in the optimized LSF.