Efficient scheduling strategies in High Performance Computing Service Platform for Shanghai Colleges

High Performance Computing Service Platform for Shanghai Colleges (HPCSP) is built to meet the mass of computing requirements of teachers and students in these colleges and universities by Shanghai Municipal Education Commission, China. HPCSP contains several high performance computing clusters, so a new efficient job scheduling strategy is required in this platform. Two level schedulers are presented in this paper, and Global Job Scheduling Algorithm (GJSA) and Local Job Scheduling Algorithm (LJSA) are given. The key feature of GJSA is taking job and node characteristics into account. LJSA is a local scheduling algorithm based on user priority and CPU number without prediction of executing time. At last GJSA and LJSA are experimentally demonstrated to reduce the Average Waiting Time of jobs and increase the utilization rate of clusters in HPCSP.