Comments on PVPs,MPPs, NOWS,and future computer architectures

Using 1995 results from the NASA Parallel Benchmarks, this paper compares the sustained performance (as opposed to peak performance) and cost per unit of sustained performance for state-of-the-art Parallel processing systems that use Micro-Processors (MPP) versus those that use Custom-designed Processors (PVP). This comparison shows that MPPs approach, but seldom surpass, PCPs with respect to either metric. This paper also discusses the merits and demerits of “clusters” of workstation. Ironically, recently introduced low-cost, parallel systems can also be viewed as “virtual workstation clusters” that offer comparable, if not superior, capabilities. Finally, this paper explores the question of whether it will be economically feasible for computer manufacturers to continue building PVPs, or are MPPs the only economically viable option for building parallel systems?