Research On Job Scheduling Algorithm Based On Pc Cluster

According to the data of Top500, cluster system has been the mainstream of high-performance computing system. And the job management system is the key of cluster. Job scheduling algorithm is the core of job management system and the most important part of the whole system. A good job scheduling algorithm not only reduces the waiting time of a job, but also makes full use of resources in system, and improves CPU utilization. OpenPBS is one of the most popular job management systems, which has a good support for high-performance computing. Backfilling algorithm as an effective scheduling algorithm has been more and more recognized.This paper mainly focuses on job management system and the scheduling algorithm, elaborates the advantages and disadvantages of OpenPBS system and backfilling algorithm. The FIFO scheduling algorithm of OpenPBS may cause some idle resources in the system, and when the idle CPUs can't meet the requirements of the smaller backfilled jobs, there are still idle CPUs, and the CPU utilization can't be better enhanced. In order to resolve those problems, LA-RB (Large Area-Reservation Backfilling) is proposed in this paper. It is proved reasonable and effectiveLA-RB algorithm is based on largest area first. The first scheduled large area jobs improve the job fairness. Introducing two level priority and Reservation Algorithm can eliminate the hunger of bigger jobs, reduce the running time of jobs and enhance the CPU utilization.In the paper, a cluster with 10 nodes was built. The tested result is that reservation-backfilling algorithm can bette??prove the job fairness, CPU utilization and the average response time than the traditional backfilling algorithm.