Power Profiling: Understanding the Impact of CPU Workloads on Container and Virtual Machine Power Efficiency

Abstract: Data centers power consumption represent a substantial amount of the global power consumption. While hardware has improved over the years, this study focuses on the software side of optimization, looking at the power consumption differences in containers and virtual machines.Through testing, this project reveals that there is a significant difference in the power consumption of containers and virtual machines. The type of application load imposed on the CPU is a critical factor that significantly changes the power consumption characteristics of the virtualization technologies.In purely CPU-bound tasks, containers exhibit a marginal edge, in being 0.6% more efficient than virtual machines. When syscalls are being invoked, the results show that virtual machines are the more energy efficient choice by a margin.As the workload tends more towards that of I/O operations, the containers are once again the most energy efficient option. Even though this study shows that one can reduce power consumption by replacing virtualization technologies with eachother, more prevalent methods exists that reduce the power consumption even further than what this project has managed.