Load-Balancing and Prudent Deployment of VNFs for Heterogeneous Multicore Systems

Network function virtualization (NFV) is an enabling technology for telecommunication operators to achieve flexible service deployment and efficient reuse of physical resources; operators are no longer restricted to dedicated hardware and can run their services on off-the-shelf servers. However, there has been little research effort on optimizing NFV for heterogeneous multicore architectures such as ARM's big.LITTLE technology and Intel's Performance Hybrid Architecture. In this paper, we consider CPU core affinity of virtual network functions (VNFs) and study load-balancing and prudent deployment of VNFs in a heterogeneous multicore system. We not only develop the Four-Class Filling (FCF) algorithm with a complexity of o(n log n); and an approximation ratio of 3/4, but also implement it on an NFV platform equipped with two types of CPU cores. Real-world measurement validates feasibility and effectiveness of our method. Besides, simulation results show that our algorithm performs very well in terms of load-balancing factor, operational cost, and excess usage.