Assessing the Power Cost of Virtualization Through Real-world Workloads

Abstract

Next-generation mobile networks will be heavily based on virtualization and their pervasiveness raises many questions regarding the energy efficiency of an architecture that requires distributed computing resources at the network edge. In this paper, we focus on the two main virtualization approaches, i.e., virtual machines and containers, which play a primary role in the provisioning of MEC-based services for mobile users. Specifically, we compare the two approaches from the viewpoint of the power consumption they are associated with - a metric significantly affecting the network provider's costs as well as the ICT environment footprint. Through a set of realworld experiments, using real-world video streaming and gaming applications, we assess not only the magnitude of the power consumption we incur, but also how it evolves as the workload increases. Our results show that containers are both more powerefficient and more scalable than virtual machines.