Resource Management and Scheduling on SupernodeII

In recent decades, the demands of using computer to solve grand and challenging problems grow both in size and in complexity. Distributed and parallel computing is thus important. Enabling technologies in highspeed communication today have made PC-based clusters become a mainstream of parallel and distributed platforms for high-performance, high-throughput and high-availability computing. To enable effective resource management on clusters, numerous cluster management systems and schedulers have been designed. In clusters, the processors can be shared in two basic ways: space-sharing and time-sharing. In this paper, the anatomy of a resource management and scheduling system for SupernodeII – a high-throughput 64-nodes PC-based cluster – is shown. The system uses space-sharing mode. Relying on the “adaptive partitioning” approach to allocate nodes to jobs, we propose some improvements on the approach so that it can be applied to SupernodeII. Optimizing system utilization, reducing resource fragmentation, preventing indefinite postponing, getting fairshare among jobs are important results we achieve.