Revisiting NFV orchestration with routing games

Network Function Virtualization (NFV) is an emerging approach that has received attention from both academia and industry as a way to improve flexibility, efficiency, and manageability of networks. This flexibility can be leveraged to allow for a better optimization of Virtual Network Functions (VNFs), which can now be dynamically placed in the network to adapt efficiently to both network constraints and the business objectives of operators. Network applications' requests are expressed as Service Chains (SCs): a set of VNFs chained together through virtual links. The placement of VNFs and SCs is a challenge that impacts the quality of service of all applications and the operating cost incurred on the operator. In this paper we present an Integer Linear Programming (ILP) formulation of this placement problem. Since the problem is NP-complete, we provide a heuristic based on game theory and implemented in a best response algorithm. One key benefit of our approach is that it stays provably close to the optimal, when VNF costs are linear with respect to the server charge. We also evaluate this heuristic when this assumption is relaxed and show that it stays more efficient than other state of the art algorithms.