Gilbert Sih and Edward A. Lee
Proceedings of the International Conference on Parallel Processing, pp 9-16, February, 1990.
ABSTRACT
Interprocessor communication (IPC) overhead can severely degrade the performance of parallel processing systems. This paper present a unified approach to the compile-time scheduling of precedence-constrained, communicating tasks onto arbitrarily interconnected processor networks containing dedicated communication hardware. Scheduling and routing are performed simultaneously to account for limited interconnections between processors, and shared resource contention is eliminated through the scheduling of all communications as well as computations. A new scheduling heuristic called
dynamic level scheduling is proposed, which modifies the classical list scheduling methodology to account for IPC and synchronization overhead. This technique is fast, widely targetable, and displays promising performance results.