Implementing Machine Learning models in a Kubernetes cluster for a smart monitoring system

Abstract

In order to increase the rent of resources (CPU and RAM) for nodes in a Kubernetes cluster to avoid over-provisioning or under-provisioning, we present a smart system dedicated to continuously monitoring of the infrastructure, using the power of Machine Learning in analyzing and predicting the utilization of node resources and thus guaranteeing high application availability for deployment. The models of machine learning such as Long Short-Term Memory and Recurrent Neural Networks are trained on the data collected from the cluster via Prometheus, to produce reliable predictions of future resource consumption. The results of our work are visualized in Grafana, providing a an interactive and a clear dashboard for monitoring performance. Our approach aims to improve the operational efficiency, resilience, and performance of applications running in a Kubernetes environment.