Empirical Evidence of Reliability Growth in Large-Scale Networks

An analysis of major telecommunications outages experienced by a nation-wide network is presented. The purpose of this analysis is to examine the utility of Nonhomogeneous Poisson Process (NHPP) models in characterizing large-scale network failure behavior. The analysis not only shows the suitability of this theory, but also demonstrates the reliability growth of these network services for the time period studied. Modeling network failures as a NHPP also allows the decomposition of the failure intensity into individual hazards, providing insights into failure causes. In addition, network reliability can be characterized in classical probabilistic terms. The usefulness and limitations of this technique are discussed.