Hierarchical Multi-Agent Framework for Carbon-Efficient Liquid-Cooled Data Center Clusters

2025-02-12

AI Summary

Addressing carbon emissions from cloud computing, this work tackles the challenge of jointly optimizing workload scheduling across geographically distributed data center clusters (DCCs) and facility-level cooling system efficiency. Method: We propose Green-DCC, a hierarchical multi-agent framework: an upper layer performs carbon-intensity-aware geographic workload migration with time-shifting; a novel lower-layer dual-loop deep reinforcement learning controller—integrating dynamic liquid cooling and HVAC control—employs PPO and SAC algorithms to enable synchronized, multi-DC optimization and digital twin extensibility. Contribution/Results: We release the first open-source benchmark simulation framework for sustainable computing. Driven by real-world operational traces, our approach reduces carbon emissions by 23.7% and achieves a Power Usage Effectiveness (PUE) of 1.08 versus state-of-the-art baselines, demonstrating the efficacy and advancement of cross-spatial-temporal coordinated decarbonization.

Technology Category

Machine Learning: Efficient ML / Green AIConstraint Satisfaction and Optimization: Distributed CSP/OptimizationSearch and Optimization: Distributed Search

Application Category

Systems and Infrastructure for Web, Mobile and WoT: Sustainability and carbon-aware systems for Web, mobile, and WoTEconomics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd workResponsible Web: Sustainability and climate impact of web technologies

Abstract

Reducing the environmental impact of cloud computing requires efficient workload distribution across geographically dispersed Data Center Clusters (DCCs) and simultaneously optimizing liquid and air (HVAC) cooling with time shift of workloads within individual data centers (DC). This paper introduces Green-DCC, which proposes a Reinforcement Learning (RL) based hierarchical controller to optimize both workload and liquid cooling dynamically in a DCC. By incorporating factors such as weather, carbon intensity, and resource availability, Green-DCC addresses realistic constraints and interdependencies. We demonstrate how the system optimizes multiple data centers synchronously, enabling the scope of digital twins, and compare the performance of various RL approaches based on carbon emissions and sustainability metrics while also offering a framework and benchmark simulation for broader ML research in sustainability.

Problem

• Optimizes workload and cooling in data centers
• Reduces carbon emissions using reinforcement learning
• Addresses sustainability with digital twin technology

Innovation

• Reinforcement Learning controller
• Dynamic workload optimization
• Liquid cooling synchronization