Performance modeling for power consumption reduction on SCC

Bertrand Putigny 1, 2 Brice Goglin 2, 1 Denis Barthou 2, 1
1 RUNTIME - Efficient runtime systems for parallel architectures
Inria Bordeaux - Sud-Ouest, UB - Université de Bordeaux, CNRS - Centre National de la Recherche Scientifique : UMR5800
Abstract : As power is becoming one of the biggest challenge in high performance computing, we are proposing a performance model on the Single-chip Cloud Computer in order to predict both power consumption and runtime of regular codes. This model takes into account the frequency at which the cores of the SCC chip operate. Thus, we can predict the execution time and power needed to run the code for each available frequency. This allows to choose the best frequency to optimize several metrics such as power efficiency or minimizing power consumption, based on the needs of the application. Our model only needs some parameters that are code dependent. These parameters can be found through static code analysis. We validated our model by showing that it can predict performance and find the optimal frequency divisor to optimize energy efficiency on several dense linear algebra codes.
Type de document :
Communication dans un congrès
Hasso Plattner Institute at University of Potsdam. 4th Many-core Applications Research Community (MARC) Symposium, Dec 2011, Potsdam, Germany. 2011
Liste complète des métadonnées

Littérature citée [12 références]  Voir  Masquer  Télécharger

https://hal.inria.fr/hal-00649635
Contributeur : Bertrand Putigny <>
Soumis le : vendredi 1 mars 2013 - 09:44:48
Dernière modification le : jeudi 11 janvier 2018 - 06:22:12
Document(s) archivé(s) le : dimanche 2 juin 2013 - 02:20:11

Fichier

Performance_modeling_for_power...
Fichiers produits par l'(les) auteur(s)

Identifiants

  • HAL Id : hal-00649635, version 1

Citation

Bertrand Putigny, Brice Goglin, Denis Barthou. Performance modeling for power consumption reduction on SCC. Hasso Plattner Institute at University of Potsdam. 4th Many-core Applications Research Community (MARC) Symposium, Dec 2011, Potsdam, Germany. 2011. 〈hal-00649635〉

Partager

Métriques

Consultations de la notice

407

Téléchargements de fichiers

194