Augur predicts workflow energy before execution with 84% accuracy

Scientific workflows process massive datasets, consuming significant energy and producing carbon emissions. To schedule tasks efficiently, administrators need accurate energy predictions—but existing methods only work for previously executed workflows. Augur, developed by Kathleen West and colleagues (accepted at IEEE CLOUD 2026), fills this gap by predicting energy consumption prior to execution. It profiles both the heterogeneous cluster hardware and the workflow's computational demands, using minimal historical execution data. In tests on public and private clouds, Augur achieved a median prediction error of 16.3% (±15.3%) versus Ichnos energy estimation and 18.2% (±14.7%) versus Intel RAPL.

By outperforming two state-of-the-art methods for both runtime and total energy prediction, Augur provides a robust foundation for carbon-aware scheduling. This allows data centers to assign tasks to the most energy-efficient resources, reducing environmental impact without sacrificing performance. The approach is especially valuable for new or modified workflows, where no prior energy data exists. As cloud computing grows, tools like Augur could help tech companies meet sustainability goals while maintaining throughput.