Dual-Based Link Weight Determination Towards Single Shortest Path Solutions for OSPF Networks

Abstract: An important traffic engineering problem for OSPF networks is the determination of optimal link weight system for a given objective function such as the Fortz-Thorup (FT) piece-wise linear convex function. In this paper, we show that the optimal link weights are obtained from the slopes of the FT function when approached from the dual space. This also suggests that the slopes can be adjusted to obtain better weights, for example to reduce number of equal-cost multi-paths (ECMP), i.e., to achieve single shortest paths for different demand pairs, where possible, without unduely increasing the cost. We present an iterative objective function approach where we modify the FT function with a convex (still piece-wise linear) envelope that generates additional slope values. We present results for experimental and random networks. The results show that the iterative heuristic is fairly powerful and robust and often gives unique shortest paths for most demands in the network.