Skip to Main content Skip to Navigation
Theses

Approches statiques et dynamiques pour l'optimisation de la consommation énergétique des applications de calcul à hautes performances

Mathieu Stoffel 1, 2 
2 CORSE - Compiler Optimization and Run-time Systems
Inria Grenoble - Rhône-Alpes, LIG - Laboratoire d'Informatique de Grenoble
Abstract : The High Performance Computing (HPC) field is a crucial issue, for both industry andacademics: from astrophysics to meteorology, passing by materials chemistry, fromAirbus to Total, passing by Pfizer. Computational sciences have become essential, andthis dependence implies a neverending urge for more computational power.At the time of writing, all the actors of the HPC field redouble their efforts to reach theExaScale: 10^18 operations on floating point numbers per second. Nevertheless, contraryto the previous milestones (e.g. the PetaScale), the computational power achieved by asupercomputer is not the only key performance indicator. Indeed, the first assessment ofthe electrical power consumed by an exaflopic system were way too high to be acceptable,from both economic and ecological points of view. Consequently, numerous research anddevelopment efforts aiming at making supercomputer more energy-efficient were initiatedduring the last decade.That is precisely the main topic of the work presented by this manuscript, whichincludes significant contributions to Bull Dynamic Power Optimizer (BDPO), and theconception, development, and experimental validation of Phase - Temporality Analyser(Phase-TA).BDPO is a dynamic reconfiguration tool, that is to say a daemon executed in parallel ofan HPC application, which changes the functioning frequency of the cores of the CPUs tothe workload the latter are executing. It has the distinctive feature of being completelyagnostic of both the aforementioned executed application and its execution environment,while requiring no specific configuration from the user. Using BDPO induces a 15% decreaseof the energy consumption associated with the execution of the two applications NEMO andHPCG, while maintaining the associated performance degradation under 4%.Phase-TA is designed to analyse the profile of an iterative HPC application, notablythose produced by BDPO. It detects the locally periodic behaviours, and caracterises themby infering representative patterns for the associated periodicities. What motivated thedevelopment of Phase-TA was the possibility to build a relevant and reliable predictionof the future behaviour of the executed application, so as to make the reconfigurationsperformed by BDPO more efficient. It was experimentally shown that the patterns inferredby Phase-TA are relevant representations of the periodicities featured by HPC applications,and that those periodicities are accountable for a significant part (i.e. more than twothirds) of the execution time of the aformentioned applications. Finally, the performancesof Phase-TA make it suitable for on-the-fly analysis of the profile of HPC applications.
Document type :
Theses
Complete list of metadata

https://tel.archives-ouvertes.fr/tel-03562771
Contributor : ABES STAR :  Contact
Submitted on : Wednesday, February 9, 2022 - 11:33:32 AM
Last modification on : Wednesday, July 6, 2022 - 4:22:40 AM
Long-term archiving on: : Tuesday, May 10, 2022 - 6:35:59 PM

File

STOFFEL_2021_archivage.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-03562771, version 1

Citation

Mathieu Stoffel. Approches statiques et dynamiques pour l'optimisation de la consommation énergétique des applications de calcul à hautes performances. Architectures Matérielles [cs.AR]. Université Grenoble Alpes [2020-..], 2021. Français. ⟨NNT : 2021GRALM034⟩. ⟨tel-03562771⟩

Share

Metrics

Record views

96

Files downloads

41