Dynamic Optimization of OSPF Weights using Online Simulation

Abstract

Abstract — In this paper, we address the issues associated with the dynamic optimization of OSPF weights. For this work we have chosen the packet loss rate in the network as the optimization metric as it is a good indicator of congestion in the network and also impacts the performance of the underlying applications. The packet loss rate has been formulated in terms of the link parameters, such as bandwidth and buffer space, and the parameters of the traffic demands. A GI/M/1/K queueing model has been used to compute the packet drop probability on a given link in the network. We have developed a fast adaptive random search algorithm to address the issue of speed in the dynamic optimization problem. The proposed search algorithm has been compared to the local search algorithm of [6] in terms of the number of iterations needed to obtain a “good ” link weight setting. Both the piece-wise linear optimization function from [6] and the packet loss rate in the network have been considered as the performance metrics. Our results demonstrate that the adaptive random search takes 50-90 % fewer iterations as compared to the local search [6]. The amount of improvement depends on the network topology, traffic and the optimization metric. We have also demonstrated improvements of the order of 30-60 % in the total packet loss rate in the network by dynamically optimizing OSPF weights. Keywords—Traffic Engineering, OSPF, Optimization I.