Modelling of the parallel resolution of a numerical problem on a locally distributed computing system

• Authors: Brigitte Plateau;Andreas Staphylopatis
• Venue: SIGMETRICS '82 Proceedings of the 1982 ACM SIGMETRICS conference on Measurement and modeling of computer systems
• Year: 1982

Modern VLSI technology has enabled the development of high-speed computing systems, based upon various multiprocessor architecture [1]. We can distinguish several types of such systems, depending on the control policies adopted, the interprocessor communication modes and the degree of ressource-sharing. The efficiency of parallel processing may be significant in various areas of computer applications; especially, large numerical applications, such as the solution of linear systems and differential equations, are marked by the need of high computation speeds. So, the advance of parallel processing systems goes together with research effort in developping efficient parallel algorithms [2]. The implementation of parallel algorithms concerne the execution of concurrent processes, assigned to the processors of the system, which communicate with each other. The syncronization needed at process interaction points implies the existence of waiting delays, which constitute the main limiting factor of parallel computation. Several modelling techniques have been developped, that allow the prediction and verification of parallel systems performance. The two general approaches followed concern deterministic models [3] and probabilistic models. The latter, based on the theory of stochastic processes [5]... are well adapted to the analysis of complex variable phenomena and provide important measures concerning several aspects of parallel processing.