Timesharing on a Massively Parallel Computer

点击翻译按钮获取中文摘要

Present Massively Parallel Computers (MPCs) have been used as stand alone single user systems with a batch job interface. However, the next generation will be sufficiently powerful that time shared control systems will be required to support multiple users and minimize loss of CPU cycles. For MPCs with Multiple Instruction Multiple Data (MIMD) architectures, the dynamics scheduling of clusters of CPUs can allow effective use without idling CPUs from lack of work. A dynamic job mix control system will be developed which will optimize job throughput by maximizing the number of CPUs assigned to jobs.