In March, PhD candidate and research associate Chiara Fusar Bassini from the Data Science Lab and the Centre for Sustainability at the Hertie School, in collaboration with Leonard Hackel from Technische Universität Berlin and the Berlin Institute for the Foundations of Learning and Data (BIFOLD), and Thorren Kirschbaum from Helmholtz-Zentrum Berlin and Freie Universität Berlin, published a paper in The Weizenbaum Journal of the Digital Society (WJDS).
The paper titled “A Project Concept for AI-Assisted Energy Efficiency in HPC Clusters” introduces IDLEWiSE, an intelligent decision tool designed to reduce energy consumption in High-Performance Computing (HPC) clusters by selectively shutting down idle computational units when clusters are underused.
“The energy demand for high-performance computing (HPC) is growing, and so are concerns over its environmental impact,” explains Chiara. “In our paper, we examine the energy consumption of HPC clusters, explore strategies for reducing this consumption through efficient management tools, and discuss the role AI can play in addressing this challenge.”
The study presents an optimisation tool that uses machine learning algorithms, such as decision trees, to develop online shutdown policies for HPC systems. It also provides an analysis of existing energy-saving tools and strategically validates the proposed solution. Furthermore, the research draws insights from a survey of German scientific HPC administrators, highlighting the growing need for energy-efficient practices in the field.
To read more about this cutting-edge research, click here to access the full paper.
-
Chiara Fusar Bassini, Doctoral Programme in Governance 2023
-
Aliya Boranbayeva, Associate Communications and Events