A static scheduling system for a parallel machine (SM)2-II

Abstract

(SM)2-II (Sparse Matrix Solving Machine II) is a large scale multiprocessor for widespread use in scientific computation. In this machine, problems are described in a concurrent process language. (SM)2-II is designed so as to manage a large number of small concurrent processes effectively. However, if the granularity of the processes is fine and the number of processes becomes great, the overhead of process switching and communication bottlenecks the machine. If these processes are statically scheduled before execution, the overhead of process control can be greatly reduced. In this paper, a static scheduling system for concurrent processes is proposed. Using this system, processes are scheduled statically and merged into a smaller number of processes, according to the number of processing units. In the most successful case, no operating system is necessary. In general, exactly optimized scheduling is an NP-complete problem. In our system, a practical optimization algorithm named LS-M (Level Scheduling with Merging) is utilized. Using LS-M, semi-optimized results can be obtained within a practical execution time. Some examples, including ordinary differential and simultaneously linear equations and expert systems written in OPS5, are used to evaluate the scheduling system.