Cloud Stateless Server Failover Prediction Using Machine Learning on Proactive System Metrics

Abstract

Cloud computing, revered for its extraordinary scalability and elasticity, has revolutionized business operations by providing flexible resource options based on demand. However, this on-demand resource allocation poses distinct challenges. Due to the fluid nature of resource allocation and load distribution in the cloud, monitoring the health of servers using system metrics becomes problematic. This complexity can lead to unexpected server request failures and service interruptions due to resource insufficiencies, highlighting the need for more effective monitoring systems. Our research utilizes machine learning techniques to predict cloud server health based on resource usage and operational metrics, focusing specifically on stateless applications. Our study reveals that a Logistic Regression model trained on these system metrics delivers the most precise predictions. After hyperparameter tuning, the model exhibited robust performance, achieving a macro-averaged F1 Score of 97.7%. The paper outlines our methodology, findings, and the potential of this approach for cloud server health prediction.