Parallel Computer Architectures

Abstract

Publisher Summary The term “parallel processing” designates the simultaneous execution of multiple processors to solve a single computational problem cooperatively. Parallel processing has the potential for making difficult computational problems tractable by significantly increasing computer performance. Two basic kinds of computational problems are encouraging research in parallel processing through their need for orders-of-magnitude improvements in computer processing speed. First, problems characterized by inordinate size and complexity that often require hours or days of conventional processing. Secondly, real-time problems that requires the computations to be performed within a strictly defined time period and are typically driven by external events. These computational problems call for vast performance increases that conventional, single-processor computers are unlikely to provide. The primary barriers for achieving this kind of performance improvement through parallel processing; however, are conceptual ones, such as finding efficient ways to partition a problem among many processors and to orchestrate multiple processors executing in a cooperative fashion. Parallel processing must be supported by architectures that are carefully structured for coordinating the work of many processors and for supporting efficient interprocessor communications. The chapter discusses the major classes of parallel architecture—synchronous architectures, multiple instruction streams, multiple data streams (MIMD) Architectures, and MIMD execution paradigm architectures—describing their structure and how they function.