A Genetic Approach to Continuous Optimization of Virtual Network Embedding

Obtaining the optimum configuration for a virtual network to be embedded on a substrate network is known to be unfeasible and intractable for large networks. This limitation can be overcome by using evolutionary algorithms guided by heuristics, such as genetic algorithms. Although they are fast to reach a good configuration, it is usually just a local optimum that could be easily improved with more computation time. In this paper we propose an algorithm that, after providing a configuration in a very reduced time boundary, continues its work to get the best configuration possible within some constraints of time, number of iterations, and distance from the ideal solution. We demonstrate that, after some additional iterations, the algorithm obtains a configuration that is 7 times better than the initial configuration. Although the latter can be already enforced in the network, the improved configuration will be enforced when it is ready, so the network efficiency will be continuously improved.