Diagnosability and diagnosis of algorithm-based fault tolerant systems

Multiprocessor architectures are now in common use for signal processing and other computation-intensive applications. These applications are characterized by high-speed data processing and/or long processing periods. It is therefore desirable that any erroneous data produced by the system be detected as quickly as possible, and that the faulty processors producing the erroneous data be located and reconfigured out of the system. Algorithm-based fault tolerance (ABFT) is a low-cost, system-level concurrent error detection scheme that can also be used for locating faulty processors. Graph-theoretic and matrix-based models have been developed for the analysis of systems using ABFT. These models are used to analyze a system for its fault diagnosability. Methods used in the analysis of multiprocessor systems using system-level diagnosis are applied to the analysis of ABFT systems. Using these methods, an improved diagnosability algorithm is provided. An efficient diagnosis algorithm for ABFT systems for identifying the faulty processors, if any exist, from the information available is given. No such algorithm was known before