Links

GitHub

Tags

#Cloud and Edge Computing #Compute Frameworks

GreenFaaS

Computational scientists have access to a variety of different types of machines distributed across many sites. Typically users run their applications only using a single (typically homogeneous) machine. We find that this could be detrimental for energy efficiency. When measuring the marginal energy consumption of different functions across machines, we see a tremendous variation. Not only does the most efficient function vary by machine, but the most efficient machine also depends on the function being run, and the occupancy of the machine. This complexity makes it extremely difficult for users to run applications in an energy efficient manner. To address these issues, we propose GreenFaaS, a framework built on top of Globus Compute that allows users to monitor their energy consumption and schedule tasks in an efficient manner.

Publications

A. Kamatar et al., “Enhancing Energy Efficiency with Multi-Site Scheduling Strategies,” 2024 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW), San Francisco, CA, USA, 2024, pp. 1175-1177, doi: 10.1109/IPDPSW63119.2024.00197.

Kamatar, A., Hayot-Sasson, V., Babuji, Y., Bauer, A., Rattihalli, G., Hogade, N., Milojicic, D., Chard, K. and Foster, I., 2024. GreenFaaS: Maximizing Energy Efficiency of HPC Workloads with FaaS. arXiv preprint arXiv:2406.17710.

Funding and Acknowledgements

Available soon.

People

Alok Kamatar
Ian Foster
Kyle Chard
Valerie Hayot-Sasson