Large-scale telephone network simulation: discrete event vs. steady state
Motivated by the need of the emergency response planning community to obtain timely and accurate simulation-based predictions of the follow-up effects of telecommunications infrastructure outages, we study the trade-offs, advantages, and disadvantages of a high-fidelity time-dependent Public Switched Telephone Network (PSTN) simulator, called MIITS-P and its steady-state challenger, called IEISS. While IEISS computes results for individual simulation runs in a few seconds, MIITS-P call-by-call level simulation takes up to two magnitudes longer to complete runs, resulting in response times of tens of minutes. We study the quality of the results produced by both simulators for single-contingency cases, where we remove a wirecenter node from the network for an entire day and see what effect its removal has on the ability of the network to route its call traffic. Against our intuition that is based on a few theoretical examples suggesting that IEISS may not be able to even get base line traffic correct, we find that IEISS actually predicts relative rankings of wirecenter nodes extremely well on practical instances. This positive result holds under a specific transformation of the time-dependent simulation model into a steady-state equivalent; we look at a variety of such transformations and identify the best one. In an attempt to explain the good performance of IEISS, we take a closer graph-theoretic look at our network model, which reveals that most highly ranked wirecenters are connected in such a way that their removal cuts all possible paths between a large set of source and destination wirecenters. These cuts make time-dependencies and load variations over the course of a day insignificant and thus IEISS's steady state approach sufficient.