PhD seminar by Mats Petter Wallander, Lund university
Tid: 12/6 2009, kl 13.00
Lokal: E:1406, E-huset, LTH
Titel: Constraint Programming Methods for Optimization of Single Shortest Path Routing
Opponent: Professor Nicolas Beldiceanu, Ècole des Mines de Nantes, France
In this thesis, we propose methods based on constraint programming
(CP) for solving an optimization problem in telecommunications, the single shortest path routing problem. The problem is to find optimal values for a set of routing configuration parameters in a shortest path routing protocol, for a given network. With optimal parameters, traffic is routed through the network in a way which is optimal w.r.t.
the available bandwidths of the network links. The routing parameters define a link weight for each network link, and the routing protocol makes sure that traffic between pairs of nodes in the network is sent along shortest paths w.r.t. the chosen link weight parameters. An example of such a routing protocol is the widely used Open Shortest Path First (OSPF) protocol. We consider the version of shortest path routing that requires the shortest path to be unique for every node pair.
Our CP models for this problem use two kinds of constraints. First, we propose several different constraints for enforcing the shortest-path property, known as admissibility, on the paths used in the routing.
These constraints use new problem-specific propagation methods for ruling out inadmissible sets of paths. Our models also include resource-based constraints, for modeling the bandwidth consumption on links. Here, our central resource constraint is based on a linear programming (LP) relaxation of the problem, so our method can be understood as a hybrid CP-LP method.
In the last part of the thesis, we consider different methods for organizing the search process, and present experimental results from comparing different versions of our CP models.