Evaluation of Mechanisms for Fine-Grained Parallel Programs in the J-Machine and the CM-5

Experimental and commercial parallel machines have matured to a point where it is possible to quantify the performance enhancement due to the novel mechanisms supporting fine-grain parallel programs in the experimental machines. The MIT J-Machine provides a register-level message send, autonomous message receive and dispatch, a prioritized scheduler with multiple contexts, and tagged memory with fast traps. The Thinking Machines CM-5 is essentially a collection of conventional processors with user-level messages. However, it is difficult to establish a meaningful workload for a comparison that adequately accounts for compiler optimizations which specifically address the lack of available mechanisms. This paper uses an abstract machine approach in a relative evaluation of the J-Machine and CM-5. High-level parallel programs are translated by a single optimizing compiler to a fine-grained abstract parallel machine, TAM. The final compilation step is unique to each architecture and optimi..