Decentralized, closed loop, routing in telephone networks

Telephone networks, and other networks of interacting queues are studied as examples of stochastic dynamical systems, by using as model a Markov process with discrete space, driven by Poisson processes as external noise input. Control decisions are, first, whether to admit a new request into the system, and, for each admitted request, the choice of a route from source to destination. It is well known that admissions control is necessary to prevent a collapse of the system performance under very heavy load. The use of alternate routes leads to a more efficient use of the network capacity, but, especially with nonhierarchical routing, increases the risk of a collapse of performance under overload.