A graph approach to placement of Service Functions Chains

Abstract

Network Functions Virtualization (NFV) is a new network architecture concept which leverages virtualization technologies to make the management of network functions like firewalls, load balancers, WAN optimizers more flexible and cost effective. In this approach, traditional middlebox appliances are replaced by virtual machines embedding Virtual Network Functions (VNFs). One of the main challenges of NFV orchestration is to appropriately deploy and instantiate sequences of VNFs that form Service Functions Chains (SFC). In this paper, we propose a general cost-driven Integer Linear Programming (ILP) formulation of this problem. We then propose a graph-based heuristic that combines graph centrality and multi-stage graphs. We evaluate our heuristic by comparing to optimal solutions provided by the ILP, showing that it is 1000 times faster and very close in terms of costs (less than 1.15% of difference). Then, we evaluate our heuristic on larger instances and compare it to one of the best suited state of the art approach. The evaluation results show that our heuristic scales well and outperforms the related approach, especially when VNF deployment costs are higher than link bandwidth usage costs.