Carbon-Aware AI Workload Scheduling in Renewable Cloud Data Centers
Abstract
The rapid expansion of artificial intelligence applications has substantially increased the computational and energy demands of cloud data centers, making sustainable resource management an increasingly critical challenge. Although renewable-powered cloud infrastructures have emerged as a promising solution for reducing greenhouse gas emissions, the intermittent nature of renewable energy and the heterogeneous characteristics of workloads make workload scheduling a complex multi-objective optimization problem. This study proposes a Multi-Objective Carbon-Aware Optimization Framework for scheduling workloads in renewable-powered cloud data centers. The proposed framework is formulated as a deterministic Mixed-Integer Linear Programming model that simultaneously minimizes operational energy cost, carbon emissions, and workload makespan while satisfying computational resource capacities, renewable energy availability, and quality-of-service constraints. The experimental results demonstrate that the proposed framework reduces total carbon emissions by approximately 13.4% compared with the baseline cost-oriented scheduling policy while requiring only a moderate increase of 15.5% in operational energy cost and maintaining acceptable workload completion times. The obtained Pareto frontier provides cloud operators with multiple scheduling alternatives that effectively balance economic efficiency, environmental sustainability, and computational performance. The proposed framework offers a practical decision-support tool for next-generation renewable-powered cloud data centers and establishes a solid foundation for future sustainable cloud resource management.