Extended behavioral decomposition for estimating ultrahigh reliability

Limitations of some analytic techniques in approximating the reliability of life-critical electronic systems are discussed, and a framework for the specification of recovery and fault- handling submodels is suggested. The framework makes full use of the instantaneous jump theorem by viewing the collection of interfering, premature exits from any fault handling and recovery submodel as defining a new, competing process submodel. This approach allows a greater flexibility in submodel representation, since submodels may contain arbitrary entrance arcs, exit arcs, and competing, interfering transitions with arbitrary destinations. Since the effects of near-coincident faults need not be represented as system failure events, reliability estimates produced by this approach need not be unduly conservative. Comparisons on small models, where exact results can be computed, show substantial improvement in accuracy over earlier techniques. Implementation of the technique in an X Windows-based system, XHARP, is described. The dual top-down/bottom-up interface of XHARP provides added flexibility by allowing an automated behavioral decomposition that is based on the suggested framework.<
>