Utilisation efficace des accélérateurs GPU -- Ordonnancement sur machines hybrides

Raphaël Bleuse 1, *
* Corresponding author
1 MOAIS - PrograMming and scheduling design fOr Applications in Interactive Simulation
Inria Grenoble - Rhône-Alpes, LIG - Laboratoire d'Informatique de Grenoble
Abstract : The race for ever more computing power raises the issue of supercomputers' power consumption. Heterogeneous architectures - composed of processor and GPU accelerators - seem to be a promising answer. Scheduling on such machines is nowadays relying on the programmer's skill set and made in a static way. We study the problem of scheduling of independent tasks on such architectures. We propose a bi-objective approximation algorithm which simultaneously optimizes the makespan and the affinity. The provided algorithm has a low complexity. We then validate the performance of its implementation within the framework XKaapi.
Complete list of metadatas

Cited literature [33 references]  Display  Hide  Download

https://hal.inria.fr/hal-00858233
Contributor : Raphaël Bleuse <>
Submitted on : Thursday, September 5, 2013 - 7:41:47 AM
Last modification on : Saturday, February 2, 2019 - 1:25:42 AM
Long-term archiving on : Thursday, April 6, 2017 - 3:21:50 PM

File

rapport.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-00858233, version 1

Collections

Citation

Raphaël Bleuse. Utilisation efficace des accélérateurs GPU -- Ordonnancement sur machines hybrides. Calcul parallèle, distribué et partagé [cs.DC]. 2013. ⟨hal-00858233⟩

Share

Metrics

Record views

331

Files downloads

587