Scalable energy-aware VM allocation on cloud data centers through mathematical programming models

Abstract

Cloud data centers are becoming indispensable pillars of modern society, driving AI innovation, global connectivity, and data-driven advancements. As their size and complexity grow, so does the urgency for sustainable and efficient solutions to address operational and environmental challenges. The Virtual Machine (VM) allocation problem lies at the heart of these challenges, directly impacting energy consumption, scalability, and cost-effectiveness. While heuristics are traditionally favored for their fast execution times, they fail to adequately address the complexities of heterogeneous environments and the increasing need for energy-aware solutions. In this work, we redefine the potential of mathematical programming models — traditionally considered impractical due to scalability limitations — by defining a comprehensive VM allocation strategy that embeds the models into scalable algorithms that distribute computational workloads and exploit solver capabilities. This approach achieves linear scalability — an unprecedented milestone for mathematical programming — allowing us to integrate detailed and heterogeneous aspects of the VM allocation problem. The resulting algorithms dramatically outperform state-of-the-art heuristics and metaheuristics in both scalability and solution quality, delivering an average 16% increase in Net Profit, a 54% reduction in Total Energy Consumption, and a more-than-double improvement in Energy Efficiency. Designed to meet the evolving demands of modern Cloud data centers, our algorithms scale efficiently to manage growing workloads, adapt to heterogeneity, and comply with sustainability and regulatory requirements by prioritizing energy efficiency, facilitating the transition to next-generation Cloud environments.