A Multi-Objective Reinforcement Learning Framework Balancing Operational Costs, Carbon Footprint, and SLA Compliance in Heterogeneous Multi-Cloud Architectures

• Abiodun Okunola

  Ladoke Akintola University Technology

  Author

The proliferation of heterogeneous multi-cloud architectures has created unprecedented challenges in resource orchestration, where operators must simultaneously minimize operational costs, reduce carbon emissions, and maintain strict Service Level Agreement (SLA) compliance. Existing scheduling approaches predominantly optimize single objectives or employ static heuristics that fail to adapt to dynamic workload conditions and spatiotemporal variations in carbon intensity. This study introduces GreenBalance, a novel Multi-Objective Reinforcement Learning (MORL) framework that formulates cloud resource allocation as a Multi-Objective Markov Decision Process (MOMDP) and learns Pareto-optimal policies across three competing objectives. The framework integrates real-time carbon intensity signals, dynamic pricing data, and SLA constraint monitoring into a unified reward function that enables preference-conditioned decision-making. Experimental evaluation using production workload traces from heterogeneous cloud environments demonstrates that GreenBalance achieves 89.4% SLA compliance while reducing operational costs by 31.2% and carbon emissions by 26.7% compared to static threshold-based baselines and single-objective reinforcement learning approaches. The framework exhibits robust generalization across diverse workload patterns and cloud provider configurations, establishing a replicable methodology for sustainable multi-cloud orchestration. The findings have significant implications for practitioners seeking to balance economic and environmental objectives, and for researchers advancing multi-objective reinforcement learning in distributed systems.

• pdf

Copyright (c) 2026 Abiodun Okunola (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.