Reconfiguration capability in asynchronous transfer mode (ATM) networks and its ability to reduce ATM cell losses are studied. An implementation that restricts the transmission of ATM cells of an O-D pair to certain channels in each physical link in a simple path from origin to destination is proposed. The joint topology, discrete capacity, and routing problem in a reconfigurable ATM network is formulated as a nonlinear mixed integer programming problem. It is proven that this problem with integrality constraints relaxed can be simplified to a convex programming problem, providing a method for computing a lower bound on the original problem. An algorithm is developed to solve the nonlinear mixed integer programming problem. In computational experiments, the reconfiguration capability decreased the total cell losses by more than 50%.>

A flow model for evaluating the performance of a network of asynchronous transfer mode (ATM) switches is presented. The performance measures used include the link (nodal) and end-to-end cell loss probabilities as well as the link (nodal) and end-to-end cell delays. In the model, the routing assignments are assumed to be given. The assumed form of routing assignments may be used to represent either virtual circuit or datagram service. Due to the nonlinear relationship between cell losses and offered flows, the flow model is a system of nonlinear equations. The authors develop a sufficient condition for the existence of a unique solution to the nonlinear system of equations. They present an iterative model and prove that it converges to a unique fixed point provided the sufficient condition is satisfied. The unique fixed point corresponds to the unique solution to the flow model.>

The configuration and routing problem (or the joint trunk and traffic routing problem) is first formulated as a mixed integer nonlinear minimization problem. Then certain properties are proved about the problem structure. These properties lead to an efficient algorithm that determines near-optimal configurations and routing assignments. A lower bound is also computed in order to evaluate the quality of the solution. The algorithm was coded and tested on several networks. Present computational experiments clearly show that the performance of a self-planning network is always better than that of the same network without the trunk routing capability. The reduction in average packet delay is as much as 67%.>