An Efficient Kernel-level Scheduling Methodology for Multiprogrammed Shared Memory Multiprocessors

In this work we present an innovative kernel-level scheduling methodology designed for multiprogrammed shared-memory multiprocessors. We propose three scheduling policies equipped with both dynamic space sharing and time sharing, to ensure the scalability of parallel programs under multiprogramming while increasing processor utilization and overall system performance. Our scheduling methodology is designed for multidisciplinary multiprocessor schedulers that need to handle applications with widely different resource requirements and execution characteristics. We implemented the policies on a 64-processor SGI Origin2000 running Cellular IRIX and evaluated them in direct comparison with the native kernel scheduler. Our results demonstrate solid performance improvements over the vendor execution environment.