UC San Diego

GreenGrader is a carbon-aware distributed autograder system designed to minimize the environmental impact of computational workloads. Autograding, the automated assessment of student assignments, is increasingly utilized in computer science education. While valuable pedagogically, it can be resource intensive. GreenGrader aims to minimize the carbon footprint of these workloads through energy-efficient computing and carbon-aware scheduling. It consists of an ingestion pipeline to receive submissions and an execution pipeline to evaluate them using containers across distributed infrastructure. By integrating with a carbon-aware scheduler and the National Research Platform’s HyperCluster, GreenGrader enables geographic workload shifting to optimize carbon emissions. The efficacy of GreenGrader was evaluated using 134 genuine student submissions. Compared to static geographic placement, GreenGrader reduced carbon emissions by 40.91% by shifting workloads based on real-time carbon intensity data, demonstrating the promise of carbon-aware scheduling. Overall, GreenGrader represents an advancement in aligning distributed computing with ecological stewardship. As society advances towards low-carbon systems, GreenGrader provides a model for embedding environmental responsibility within computational workloads.