A branch-and-bound approach to the optimization of redundant software under failure correlation

A technique for computing optimal redundancy levels for the components of a fault-tolerant system with dependently failing redundant module versions is presented. Optimization is done under a given budget constraint and with the overall reliability as the objective. Within each component, price and estimated reliability of each redundant version is assumed to be constant. For solving the optimization problem, a branch-and-bound technique, using the solutions of continuous problem relaxations for the determination of bounds, is developed. Computational experiments with randomly generated test instances show that the suggested algorithm still performs well in situations where a computation based on dynamic programming is not feasible anymore.