| Past methods to compute the reliability of a network have been based on the assumption that the components of the network fail independently of each other. The present approach discards the assumption of independence, which is often unrealistic, and seeks best lower and upper bounds for various reliability measures on a network when bounds on the failure probabilities of the network components are known.;A general framework for reliability analysis is developed in terms of blocking systems. It is shown that polynomial algorithms exist in some cases.;Tightening of the bounds is achieved by introducing into the model bounds on the probabilities that pairs of network components fail. A heuristic algorithm is developed to determine the tighter bounds on the probability that two distinguished vertices communicate in a graph. Computational results are presented on this extended model. |
