An Online Placement Scheme for VNF Chains in Geo-Distributed Clouds

Abstract

Network Function Virtualization (NFV) provides virtualized network services through service chains of virtual network functions (VNFs). VNFs typically execute on virtual machines in a cloud infrastructure, which consists of geo-distributed cloud data centers. Compared to traditional cloud services, key challenges in virtual network service provisioning lie in the optimal placement of VNF instances while considering inter-VNF traffic and end-to-end delay in a service chain. The challenge further escalates when a service chain requires online processing upon the its arrival. We propose an online algorithm to address the above challenges, while aim to maximize the aggregate chain valuation. We first study a one-time VNF chain placement problem. Leveraging techniques of exhaustive sampling and ST rounding, we propose an efficient one-time algorithm to determine the placement scheme of a given service chain. We then propose a primal-dual online placement scheme that employs the one-time algorithm as a building block to make decisions upon the arrival of each chain. Through both theoretical analysis and trace-driven simulations, we verify that the online placement algorithm is computationally efficient and achieves a good competitive ratio.