Theoretical Performance-Based Cost-Effectiveness of Multicomputers

The CPU cycles that are stolen to relay messages can significantly affect the performance of a multicomputer system. This degradation in performance in turn affects the overall cost-effectiveness of such a system. This paper compares the cost-effectiveness of four multicomputer architectures that have received a great deal of recent attention: the bidirectional straight line, the unidirectional ring, the square mesh, and the binary hypercube. Cost-effectiveness is measured by finding the ratio of the total dollar cost of a multicomputer, based upon the total cost of processors and communication links, to the total potential utilization of the processors making up the system, where the potential utilization is the fraction of time a processor has to do useful processing after subtracting time spent handling messages. The smaller the value of this ratio, the greater the cost-effectiveness of the system. 1. INTRODUCTION With the current widespread interest in parallel and distributed processing, the subject of multicomputers has received considerable attention in literature. This interest is quite natural when one considers the ever increasing demands that are being placed upon computer systems and the fact that there appears to be a fundamental limit to the computing power that can be compressed upon a single chip. An early taxonomy that categorizes systems of interconnected computers based upon strategy of message transfer, transfer control, path structure, and specific system architecture, is given by Anderson and Jensen. 2 In a survey by Feng, 6 the space of interconnection networks is represented as the cross-product of sets of design decisions including operation mode, control strategy, switching methodology, and network topology. The problem of optimizing the performance of multicomputer systems has been addressed by a number of researchers. Three common sense factors including the interconnection topology, the scheduling and mapping of the algorithm to the topology, and parallelism detection are pointed out by Agrawal and JanaKiram.