A probabilistic method for fault diagnosis of multiprocessor systems

The authors present a system-level fault-diagnosis algorithm for identifying faulty and fault-free units in a homogeneous system of computing elements. The algorithm is based on a comparison approach where tasks are performed by the units and their outputs are compared among themselves. Unlike other approaches, the authors´ algorithm requires no global syndrome analysis and therefore can be performed in real time as a background task during system operation. The time required to perform the diagnosis is constant regardless of the number of units in the system. Like previous global syndrome-based approaches, the accuracy of the algorithm is remarkably high, since it uses information about individual comparison results which is lost when these results are summarized in a global syndrome.<>