Integrated error processing and fault diagnosis in multiprocessor systems

This paper deals with Multiprocessor Systems required to provide both high performance and good figures of dependability attributes, and where processors are affected by permanent, intermittent and transient faults. Fault tolerance is pursued through a proper combination and integration of fault treatment and error processing. Our approach to fault treatment is based on a diagnostic mechanism that, accounting for transient faults, allows to ba1ance between two conflicting requirements. The first is to maintain in the system processors that have experienced just transient fau1ts and the other is to quickly remove those affected by permanent or intermittent faults. The delay in removing faulty processors and the probability of erroneously removing a non faulty processor exhibited by this diagnostic mechanism are also evaluated. Then the fault treatment is integrated with a set of error processing approaches and the complete fault tolerance strategy is evaluated by meansof stochastic simulation.