A multi-deme parallelization of a memetic algorithm for the weight setting problem in OSPF and DEFT

The Internet is divided into Autonomous Systems, which control their intra-domain traffic by using interior gateway protocols. The most common protocol used today is Open Shortest Path First (OSPF). OSPF routes traffic on shortest paths defined by integer link weights. The weight setting problem is to find weights that optimize the resulting traffic, for example to minimize network congestion. A recently proposed protocol called Distributed Exponentially-weighted Flow Splitting (DEFT) sends flow on non-shortest paths, with an exponential penalty for longer paths. Since these problems are hard to solve exactly, several heuristics have been proposed. We propose a parallel, multi-deme version of a memetic algorithm to solve the weight setting problem in DEFT. It consists of a shared memory parallelization of the (single deme) memetic algorithm, as well as instances of the memetic algorithm running in parallel, and migrating solutions among populations according to the island model. Computational results show a reduction of execution time, and an improvement of solution quality compared to the original memetic algorithm.