Automation and Multi-Objective Optimization of Virtual Network Embedding

The need for automated management is continuously increasing, especially with the advent of network virtualization and slicing technologies. However, finding the optimum configuration for a virtual network before it is embedded onto the substrate network is a problem that cannot be resolved by exact and deterministic mathematical operations. In this paper we propose a novel heuristic for building an algorithm based on genetic programming for optimizing the placement of virtual network function instances before they are deployed, so more instances can be deployed on the same substrate network without incurring in overloads and delays. Each solution given by our algorithm is based on a previous solution, following dynamic programming scheme to minimize processing and enforcement efforts. Therefore, the algorithm accomplishes with the time constraints set by current demands. We demonstrate this quality and compare our algorithm to previous solutions, also based on genetic programming and already providing quite fast responses for the embedding problem.