Performance analysis of a multiprocessor machine based on data flow principles

Abstract

In this paper we propose a model of a static data flow oriented multiprocessor system and study issues pertaining to the scheduling and communication delays with the aim of improving the efficiency of the system. We model the machine by traversing the states of processing elements during the execution of a program. We carry out performance analysis of the machine for a program P, represented by different data flow graphs. These graphs are obtained using a coalescing algorithm that keeps the loss of parallelism at a minimum. We evaluate the machine for different configurations by varying the number of processing elements and the speed of various components. The machine is modelled for three scheduling strategies, viz., dynamic, static and quasi-dynamic. In this paper we provide details of the machine using static and quasi-dynamic scheduling strategies only.