Design of multiprocessor systems for concurrent error detection and fault diagnosis

Results on the design of systems using algorithm-based fault tolerance (ABFT), a low-overhead fault tolerance scheme for high-speed parallel processing systems, are presented. Bounds on the diagnosability of the system and the number of checks needed to design a unit system of given capability are derived. A procedure for forming the target fault-tolerant system from the unit system is introduced. The procedure is applicable to a wide range of systems in which processors may share data elements. The applications of the design scheme are illustrated through examples.<>