Scheduling Parallel Job on Param Padma with Mpich

Abstract

We have implemented a job scheduling system for massively parallel system with Mpich that supports different scheduling policies for execution of parallel jobs. The system is available on C-DAC HPC cluster Param Padma. It is highly modular scalable and can easily be adapted to variety of other MPP systems. Conventional job scheduler daemons take a part of resources continuously for its own activity that reduces the overall computational metric of the cluster. The scheduler minimized this resource usage by eliminating scheduler acceptor and resource monitoring daemons on the each compute nodes. This idea provides better utilization of resources on the cluster.