Diagnostic model and diagnosis algorithm of a SIMD computer

Self-diagnosis of systems comprising large numbers of processors has been studied extensively in the literature. The APEmille SIMD machine, a project of the National Institute of Nuclear Physics (INFN)of Italy, was offered as a test bed for a self-diagnosis strategy based on a comparison model. Because of the general machine architecture and some design constraints,the standard assumptions of the existing diagnosis models are not completely ful_lled by the diagnosis support built in APEmille. This circumstance led to the development of a specific diagnostic model derived from the PMC and comparison models. The new model introduces the concept of direction-related and direction-independent faults. The consistency of this model with the APEmille architecture is discussed, and possible fault scenarios which are particularly critical for the correctness of the diagnosis are examined. It is shown that the limited hardware redundancy, extended with simple functional tests, is sufficient for obtaining valid diagnosis with the presented model.