Renewable-Aware Cooperative Scheduling for Distributed AI Training Across Geo-Distributed Data Centers

Abstract

The rapid expansion of artificial intelligence workloads has dramatically increased data center energy consumption and carbon emissions. This paper presents a renewable-aware cooperative scheduling approach for distributed AI training across geo-distributed data centers. The proposed methodology exploits spatial and temporal variations in renewable energy availability to minimize carbon footprint while maintaining training performance. A two-phase optimization framework coordinates workload placement decisions across multiple data centers by predicting renewable energy generation patterns and carbon intensity fluctuations. Experimental evaluation using real-world carbon intensity data from six geographic regions demonstrates 47.3% carbon emission reduction compared to performance-optimized scheduling, achieving 86.2% renewable utilization while maintaining 96.4% deadline satisfaction rate.