Conformal Prediction for VM Sizing in Cloud Data Centers

A new study proposes a data-driven prediction interval construction method using conformal prediction (CP) for mid- and long-term forecasting of virtual machine (VM) utilization in cloud data centers. The approach, called Right-sizing Recommendations (RSR), aims to improve resource allocation by accounting for fluctuating and unpredictable VM usage, reducing over- or under-provisioning. The research focuses on large cloud providers and hyperscalers, where efficient VM sizing is critical for cost efficiency and performance. The paper is published on arXiv (2607.24773v1) and presents a framework that leverages CP to generate high-quality interval predictions, supporting cloud operators in instance provisioning decisions.