Consider a communication network whose links fail independently and a set of sites named terminals that must communicate. In the classical stochastic static model the network is represented by a probabilistic graph whose edges occur with known probabilities. The classical reliability (CLR) metric is the probability that the terminals belong to a same connected component. In several contexts it makes sense to impose the stronger condition that the distance between any two terminals does not exceed a parameter d. The probability that this holds is known as the diameter-constrained reliability (DCR). It is an extension of the CLR. Both problems belong to the NP-hard complexity class; they can be solved exactly only for limited-size instances or specific network topologies. In this thesis we contribute a number of results regarding the problem of DCR computation and estimation. We study the computational complexity of particular cases parameterized by the number of terminals, nodes and the parameter d. We survey methods for exact computation and study particular topologies for which computing the DCR has polynomial complexity. We give basic results on the asymptotic behavior of the DCR when the network grows as a random graph. We discuss the impact that the diameter constraint has in the use of Monte Carlo techniques. We adapt and test a family of methods based on conditioning the sampling space using structures named d-pathsets and d-cutsets. We define a family of performability measures that generalizes the DCR, develop a Monte Carlo method for estimating it, and present numerical evidence of how these techniques perform when compared to crude Monte Carlo. Finally we introduce a technique that combines Monte Carlo simulation and polynomial interpolation for reliability metrics.
