Load Shared Sequential Routing in MPLS Networks: System and User Optimal Solutions

Recently Gerald Ash has shown through case studies that event dependent routing is attractive in large scale multi-service MPLS networks. In this paper, we consider the application of Load Shared Sequential Routing (LSSR) in MPLS networks where the load sharing factors are updated using reinforcement learning techniques. We present algorithms based on learning automata techniques for optimizing the load sharing factors both from the user equilibrium and system optimum perspectives. To overcome the computationally expensive gradient evaluation associated with the Kuhn-Tucker conditions of the system optimum problem, we derive a computationally efficient method employing shadow prices. The proposed method for calculating the user equilibrium solution represents a computationally efficient alternative to discrete event simulation. Numerical results are presented for the performance comparison of the LSSR model with the user equilibrium and the system optimum load sharing factors in some example network topologies and traffic demands.