A decomposition approach to solving core network design problems

Abstract

Traditionally, when automated planning is used, network planners solve multilayer core network problems in a top-down manner, solving capacities for the top-most layer, and then using these solved capacities to solve the next lower layer. This results in a suboptimal solution, yielding higher capital expenditure costs. In this work an exact multilayer network Mixed Integer Linear Programming (MILP) model is developed that integrates multiple layers into a single model. Each layer takes the form of a multicommodity flow problem. The objective is to minimize capital expenditure costs, and the integrated network model is shown to be able to reduce costs. This however aggravates the computational burden, and as such, methods to improve scalability and tractibility are developed. This is done by decomposing the problem as per Benders decomposition and applying column generation. A heuristic warm-start is also developed based on this approach. The performance enhancements are compared to an integrated arc-based formulation. Advances in Ethernet technologies have resulted in lower cost hardware, scalable interfaces and flexible packet services, and together with Wavelength Division Multiplexing (WDM) facilitates low cost per bit transmissions. In order to demonstrate the flexibility of an integrated multilayer network model, the general model is applied to Ethernet over WDM networks.