3 Parallel computer architectures

Abstract

Publisher Summary Numerous high performance computer architectures reduce the computation time of many applications, but these architectures differ significantly in their approaches. These architectures include pipelined processors, vector processors, single instruction stream multiple data stream organization (SIMD) parallel processors, and multiple instruction stream multiple data stream organization (MIMD) parallel processors. This chapter describes these architectural concepts and discusses the way the application characteristics dictate the architectural requirements and performance. Computer architectures can be broadly classified into four categories using Flynn's classification: (1) the single instruction stream single data stream organization (SISD), which represents the traditional Von Neumann architectures; (2) the SIMD, which includes most array and vector processors; (3) the MIMD, which includes most systems referred to as “multiprocessors;” and (4) the multiple instruction stream single data steam organization (MISD). To reduce the execution time of a program, the traditional Von Neumann architectures can be changed by (1) allowing the overlapping of instructions or (2) allowing an instruction to operate on multiple data streams (operands), or (3) allowing multiple instructions to execute concurrently. These three variations lead, respectively, to the three types of architectures that are discussed in the chapter—pipelined processors, SIMD parallel processors, and MIMD parallel processors.