MERSEM Framework Cuts Cloud Carbon by 12% with Evolutionary RL

Researchers from multiple institutions propose MERSEM, a novel framework that tackles the growing energy and carbon footprint of graph analytics workloads in edge-cloud systems. Graph analytics powers critical applications like smart cities, IoT security, and large-scale social networks, but its execution across heterogeneous edge-cloud environments leads to high energy consumption and carbon emissions. MERSEM integrates evolutionary search with reinforcement learning (RL) to solve workload allocation and scheduling. The evolutionary component explores diverse global solutions for resource allocation, while the RL agent refines decisions through adaptive local optimization, balancing trade-offs between performance and sustainability. The framework considers dynamic carbon intensity, resource heterogeneity, and workload characteristics to jointly minimize service-level agreement (SLA) violations and carbon emissions.

Experimental results demonstrate MERSEM outperforms state-of-the-art methods with up to 45% reduction in SLA violations and up to 12% reduction in carbon emissions. These results highlight the potential of hybrid AI approaches for sustainable computing, particularly as graph analytics workloads continue to scale in complexity. By optimizing scheduling decisions in real-time across edge-cloud systems, MERSEM offers a practical path toward greener, more reliable infrastructure for data-intensive applications. The framework's ability to adapt to changing carbon intensity and resource availability makes it especially relevant for organizations aiming to meet sustainability goals without sacrificing performance.