Towards Green Container Management: A novel approach for container resource allocation through statistical modeling

Abstract

In this exploratory study, a comprehensive framework is presented that leverages ARIMA and Facebook's Prophet time series forecasting models for predicting container resource usage in a Kubernetes environment. The framework includes data collection and processing, as well as the development of an algorithm for suggesting resource allocation, which is combined with the Prophet model. Multiple KPIs were devised to evaluate the performance of the resource allocation algorithm, the statistical models, and a combined method. The proposed framework offers a successful solution for suggesting resource requests and limits. When applied to a sample Kubernetes node, the framework resulted in a 39.3% reduction in allocated resources and a 60% increase in memory usage coverage. This approach significantly benefits Kubernetes environments, promoting greener resource management and reduced electricity costs.