A Machine Learning-Based Carbon-Aware Workload Scheduling Framework for Sustainable Cloud Computing

The fast growth of cloud computing systems has led to a major rise in global energy use and carbon emissions. Most traditional methods for scheduling workloads in the cloud focus on performance, cost savings, and efficient use of resources, but they often don’t take into account the changing levels of carbon intensity in electricity grids at any given time. This paper introduces a new framework for workload scheduling that uses machine learning to be more aware of carbon emissions, aiming to lower the environmental impact without sacrificing the quality of service. The system combines predictions about carbon intensity, workload needs, and smart scheduling through supervised machine learning techniques. It assigns workloads to data centers in real time, considering expected carbon intensity, available resources, and service level agreements. The framework is built using Python-based microservices, REST APIs, and a containerized cloud simulation setup. Testing shows that this carbon-aware scheduling approach greatly reduces emissions compared to standard methods like round-robin or cost-based scheduling, while keeping system performance consistent.