Issues and directions in scalable parallel computing

Abstract

The design and use of scalable parallel computers is discussed. Much of the knowledge gained in the study of distributd systems is relevant, although it needs to be used in new manners. The author argues that the challenge is to achieve maximum reuse of existing technologies, microprocessor architectures, I/O systems, programming languages, operating systems, application codes, without compromising too much parallel performance and scalability. Such skillful compromises are essential to the success of the parallel processing industry, and pose even more intellectual challenges than the 'start from scratch' approach.