Online VNF chain deployment on resource-limited edges by exploiting peer edge devices

Abstract

Instead of deploying network functions on proprietary and dedicated hardware, Network Function Virtualization (NFV) enables to deploy network functions on virtual instances, such as virtual machines and containers. With the advance of 5G technologies, recent studies [1, 2] have shown that NFV has great potential to provide low-cost as well as ultra-low-latency services. Unfortunately, edge devices are in general of limited resource. A deep understanding of what kind of resources network function chain consumes on such edge devices and how to deal with the resource scarcity is yet unexplored. In this paper, through real experiments on a wireless router, we observe that in addition to the computation consumed CPU, communication between consecutive network functions in a chain also consumes significant amount of CPU on edge devices. To deploy network function chains on such resource-limited edge devices, we refer to peer edge devices. Concretely, we distribute part of network functions on peer edge devices that have both close proximity and relative low load. To this end, we formulate an online VNF chain deployment problem that addresses both the delay of chains and the maximum load among all edge devices by Integer Linear Programming (ILP). We prove that this problem is NP-complete. With respect to the chain delay and the maximum load, an approximation algorithm is designed, after which an online redeploy algorithm is proposed to further reduce the maximum load among all edge devices. Extensive simulations show that the proposed online deploy algorithm reduces the maximum load by 30% ∼ 50% compared with state-of-the-art deployment algorithms in edges. The online redeploy algorithm further reduces the maximum load by about 5%.