Reliability Issues in Computing System Design

Abstract

This paper surveys the various problems involved in achieving very high reliability from complex computing systems, and discusses the relationship between system structuring techniques and techniques of fault tolerance. Topics covered include: 1) protective redundancy in hardware and software; 2) the use of atomic actions to structure the activity of a system to limit information flow; 3) error detection techniques; 4) strategies for locating and dealing with faults and for assessing the damage they have caused; and 5) forward and backward error recovery techniques, based on the concepts of recovery line, commitment, exception, and compensation. The ideas described relate to techniques used to date in systems intended for environments in which high reliability is demanded. Three specific systems, the JPL-STAR, the Bell Laboratories ESS No. 1A processor, and the PLURIBUS, are described in some detail and compared.