Analytic models of pipeline and multiprocessor computer systems

Abstract

Pipeline and Multiprocessor computer organizations potentially offer attractive computational gains along many design dimensions. This paper describes the results of an analytic study of pipeline and multiprocessor computer systems to obtain bounds on the relative performance of designs. It investigates the effects of the depth of pipeline and of multiprocessing on performance of systems. It is shown that the optimum payoff in design of pipelined machines occurs with pipes that have only a few instructions of overlap. It is demonstrated that Multiprocessor systems may be effectively constructed from structures that exhibit lesser degree of tight coupling.