Year: 2025
Type: conference-paper
Abstract: Data centre monitoring and optimisation can be done by several individual, unrelated tools, each one offering its best effort, with potential overlap but also some missing features.In the EAR project, we believe that maximum efficiency can be obtained with a unified solution.The EAR software architecture for computational elements is already used in production systems.Its architecture includes node and job monitoring, energy optimisation, and cluster powercap as main features.This paper presents the extension of the EAR software architecture for holistic data centre power management.The proposal relies on a highly extensible and configurable component: the Energy Data Centre Monitor (EDCMON).This component can be used to monitor, report, and control any computational and non-computational part of the data centre.We used it to implement thermal capping, power monitoring of noncomputational devices, and cooling system monitoring and control.We present an initial evaluation of the thermal capping extension, which shows the potential of fine-grain monitoring and control of the data centre temperatures by running a real scenario with a multi-GPU workload.
Authors: Julita Corbalan, Luigi Brochard, Jalal Lakhlili, Marco d'Amico, Oriol Vidal, Jordi Aneas
Institution: Barcelona Supercomputing Center
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Award: G7633615504
OpenAlex ID: W4411345914