Exploring resource utilization in virtualization: A comparative study across platforms and hardware configurations

Abstract

With the rapid development of cloud computing, artificial intelligence (AI), and 5G networks, ICT energy consumption has become a substantial part of global energy usage. Increasing attention is being given to technological development that advocates for environmental protection, reduces resource waste, and improves resource utilization efficiency to minimize environmental harm. This thesis explores the performance and energy efficiency of different virtualization technologies when running AI applications on various types of hardware.

The research examines the performance and energy efficiency of bare metal, Docker, KVM, and VirtualBox across different hardware configurations, including a basic laptop, and a workstation without and with a GPU. Metrics such as CPU utilization, memory usage, execution time, power consumption, and energy consumption are evaluated.

The results show that bare metal consistently delivers the best performance and the highest energy efficiency in most cases. Docker also performs similarly to bare metal. KVM and VirtualBox introduce significant virtualization overhead. Additionally, when GPUs are utilized, all platforms exhibit improved performance.