The aim of this thesis is to investigate improved modeling methods to obtain reliability, performance and performability measures for three aspects of multiple processor systems: concurrency, contention and fault tolerance. We show that by using hierarchical modeling and behavioral decomposition, combinatorial models are both powerful and easy to specify. So the first set of modeling techniques investigated are combinatorial model solutions. We develop two new algorithms to solve non-series-parallel networks and multistate combinatorial models. The first algorithm uses a Boolean algebra sum-of-disjoint products method. Due to the type of operators used in this algorithm, it is more efficient than some other sum-of-disjoint products algorithms presented in the literature. A formal proof of the correctness of the algorithm is presented. The second algorithm solves multistate models combinatorially by replacing the multistate components with equivalent binary components and then applying a form of the inclusion-exclusion formula that is well-suited in this application due to the dependencies between these binary components. Many examples of such systems are provided. The second set of modeling techniques that we develop are for model generation. Several applications are modeled: a complex state-of-the-art flight control system, called the Integrated Airframe/Propulsion System Architecture, IAPSA, an I/O network called the Advanced Information Processing System (AIPS), a multiprocessor system, Cm*, and four control systems: a nuclear power plant monitoring system, a jet engine controller, a hydraulic system and a railroad control system. These systems have a large number of components with complex interdependencies that are not easy to model accurately. One important largeness avoidance method that we use in analyzing such systems is truncation. The use of truncated combinatorial models (primarily fault trees) for simple model specification is investigated in this thesis. These applications provide us with an insight into the types of systems that are more easily modeled with fault trees with Markov models.
展开▼