Scheduling Techniques with MORL to Reduce Carbon Emissions on HPC and Cloud Platforms

Data centers contribute significantly to global carbon emissions, accounting for nearly 2% of CO2 emissions worldwide. Given the urgency of climate change mitigation, this project focuses on reducing the carbon footprint of High-Performance Computing (HPC) and Cloud infrastructures by optimizing resource management strategies. Traditional scheduling methods prioritize execution time, often neglecting energy efficiency and sustainability. Parallel and distributed platforms available today are becoming more and more \textit{heterogeneous}. Cluster and Supercomputer systems are built exploiting this heterogeneity using CPUs, GPUs, FPGAs and other accelerators. The goal in a heterogeneous environment is to utilize different resources to enable the accomplishment of a given task in the shortest possible time and lately energy consumption. This project addresses the environmental impact of different computing workloads on heterogeneous HPC and Cloud infrastructures by developing advanced scheduling techniques that optimize for both performance and sustainability. Machine learning techniques will be employed to forecast energy consumption, CO2 emissions, and renewable energy requirements, aiding data center operators in sustainable infrastructure planning. The project is a work in progress and it aims to advance in the improvement of scheduling techniques, which optimizes both execution time, energy consumption and green energy sources. These algorithms will be applied to various HPC and Cloud workloads. By embedding scheduling decisions within Reinforcement Learning frameworks, the system will dynamically adjust resource allocation in response to real-time green energy availability and workload demands. This project integrates multi-objective optimization techniques, reinforcement learning, and predictive modeling to enhance energy efficiency and sustainability while maintaining high computational performance. This is a CNPq-funded (Brazilian Government) cooperation project hosted at UFPA (Amazon Region), involving other Brazilian universities (USP, UFABC, and PUC Minas) and UC Irvine (USA). The project is a work in progress and open to contributions from interested researchers and institutions.