MOBO-NFV: Automated Tuning of a Network Function Virtualization System using Multi-Objective Bayesian Optimization

With today's dramatic rise of Internet traffic, telecommunication providers increasingly rely on Network Function Virtualization (NFV) for the fast deployment and flexible orchestration of infrastructure workloads. Characterizing and tuning to run at maximum performance and/or minimum power is a critical business goal for companies.In this paper, we implement and characterize the Virtual Broadband Network Gateway (vBNG), an important NFV workload, running on a real Intel® Xeon® server. Then, we present a fully automated approach based on Multi-Objective Bayesian Optimization (MOBO-NFV) for tuning real systems running NFV workloads. We demonstrate improvements in power and performance by up to 18% and 17% respectively. We envision that the proposed methodology and results will benefit service providers for resource planning and orchestration.