The Impact of Scalability on Energy Consumption in the Cloud

Context. Companies continue to migrate their application portfolio from on- premise to cloud. Container-based virtualization and container orchestration with the open-source system Kubernetes allows cloud consumers to take ad- vantage of fundamental cloud properties, such as rapid elasticity.

Goal. In this thesis, we aim to investigate what scaling mechanisms are available in Kubernetes, and what the direct impact of the HorizontalPodAutoscaler (HPA) is on the energy consumption of a Kubernetes cluster, in order to provide advice to a university on how their energy footprint can be reduced.

Method. We perform a documentation and literature study to shape a concep- tual framework on the topics of resource management and scaling mechanisms in Kubernetes. We conduct a feasibility study with the university, and export parameters for recreating a usage scenario of one of their applications in a con- trolled environment. Finally, we perform an experiment to assess the direct impact of the HPA on the energy consumption of the application.

Results. The HPA is shown to have a statistically significant effect (a reduction) on the energy consumption of the application by scaling the number of pods based on the application’s CPU usage, as compared to a scenario where a static number of pods is running, albeit only a small decrease of 1.2%. We conjecture that the long-term deployment of the HPA can result in a more significant effect.

Conclusions. The HorizontalPodAutoscaler is a mechanism that is out-of-the- box available in Kubernetes and is easy to deploy. In our experimental setting, we show that the deployment of the HPA can be deployed in order to reduce the energy footprint of a software application.