Conformal Prediction Improves Cloud VM Right-Sizing for Cost Efficiency.

Managing cloud infrastructure efficiently is a critical challenge for large cloud providers and enterprises, particularly in optimizing virtual machine (VM) resource allocation. Traditional methods often fail to account for the unpredictable nature of VM utilization, leading to inefficiencies like over-provisioning (wasted resources) or under-provisioning (performance bottlenecks). This research proposes Right-sizing Recommendations (RSR), an AI/ML-based framework that leverages bootstrapping conformal prediction to generate high-quality prediction intervals for cloud resource demand. By learning workload patterns and forecasting medium- to long-term utilization trends, RSR aims to provide more accurate and reliable recommendations for VM sizing. This approach significantly enhances cost-effective resource allocation in dynamic cloud environments, as demonstrated by promising forecasting results and model ranking for long-life VM candidates.