System and network reliability modeling

As digital systems play a greater role in critical applications, predicting the reliability of such systems is becoming essential. The demand for both high performance and reliability leads to complex system designs that present a difficult task for the existing modeling techniques. This work is concerned with developing practical modeling and analysis techniques to help narrow the gap between the kinds of systems that can currently be modeled and those that are currently being designed.;This thesis investigates several reliability modeling problems in the area of interconnection networks and signal processor architectures. The reliability of a general network is often computed by the reliability polynomial. As a function of one variable, the reliability polynomial forces an unrealistic assumption that either nodes or links are perfect. The need for such assumptions is eliminated by extending the polynomial to two dimensions and therefore extending the class of problems to which the reliability polynomial applies. We also develop an algorithm to approximate the reliability of the n-dimensional hypercube that significantly outperforms the reliability polynomial as well as other methods.;Fault tree models are applied to the design and analysis of an FDDI token ring network. Fault trees greatly simplify the construction and verification of an FDDI network reliability model. Whereas the reliability polynomial and fault trees are primarily analysis tools, projective geometry is used to guide the proposed alternative design of an existing signal processor that is embedded in a flight control system. Projective geometry also facilitated the Markov chain analysis of the reliability and performance of such systems. We show that this projective geometry design leads to a more reliable system exhibiting superior performance in comparison with similar existing systems.