Skip to Main content Skip to Navigation
Theses

Efficacité énergétique dans le calcul très haute performance : application à la tolérance aux pannes et à la diffusion de données

Abstract : High performance computing (HPC) infrastructures have experienced a rapid growth, particularly these last years. This growth has been driven by the increased need in terms of computational power expressed by scientists in various fields. However, their increasing size makes them important electricity consumers since they already consume several megawatts. In order to consume "less" and better", we proposed a framework which permits to choose the less consuming versions of the services before pre-executing the application. In addition, our framework relies on a smart grid in order to schedule resource reservations on these computing infrastructures. This framework, called SESAMES, is adapted to two services required in high performance computing: fault tolerance and data broadcasting. Experimental validations have shown that we can reduce the energy consumption of both services by relying on accurate energy estimations provided by SESAMES for any execution context and for any platform equipped with wattmeters. Our estimation methodology is based on a description of the execution context and on a platform calibration that consists of gathering energy measurements. Simulations have shown that the multi-criteria reservation scheduler proposed in SESAMES, simultaneously reduces the energy consumption, the financial cost and the environmental impact of these reservations, while respecting the constraints imposed by the user and the energy supplier.
Document type :
Theses
Complete list of metadatas

Cited literature [115 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00881094
Contributor : Abes Star :  Contact
Submitted on : Thursday, November 7, 2013 - 2:47:09 PM
Last modification on : Monday, May 4, 2020 - 11:37:42 AM
Long-term archiving on: : Saturday, February 8, 2014 - 9:05:12 AM

File

DIOURI_Mohammed_El_Mehdi_2013_...
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-00881094, version 1

Citation

Mohammed El Mehdi Diouri. Efficacité énergétique dans le calcul très haute performance : application à la tolérance aux pannes et à la diffusion de données. Autre [cs.OH]. Ecole normale supérieure de lyon - ENS LYON, 2013. Français. ⟨NNT : 2013ENSL0836⟩. ⟨tel-00881094⟩

Share

Metrics

Record views

935

Files downloads

3897