Overview of Distributed Linear Algebra on Hybrid Nodes over the StarPU Runtime

Runtime systems usually abstract a single node, like Plasma/Quark, Flame/SuperMatrix, Morse/StarPU,Dplasma/Parsec ... When going into harnessing cluster of nodes, how should they communicate? By using explicit MPI user calls ? By using a specific paradigm (Dplasma) ? Or can we keep the same paradigm and almost the same code, and leave the runtime system handle data transfers? We show how such a system has been sucessfully implemented on top of the StarPU runtime.

Domaines

Calcul parallèle, distribué et partagé [cs.DC]

Fichier principal

siampp14.pdf (299.15 Ko)

Origine	Fichiers produits par l'(les) auteur(s)

Marc Sergent : Connectez-vous pour contacter le contributeur

https://inria.hal.science/hal-00978602

Soumis le : lundi 14 avril 2014-13:58:38

Dernière modification le : mercredi 20 mars 2024-17:52:16

Archivage à long terme le : lundi 14 juillet 2014-11:45:29

Dates et versions

hal-00978602 , version 1 (14-04-2014)

Identifiants

HAL Id : hal-00978602 , version 1

Citer

Emmanuel Agullo, Olivier Aumage, Mathieu Faverge, Nathalie Furmento, Florent Pruvost, et al.. Overview of Distributed Linear Algebra on Hybrid Nodes over the StarPU Runtime. SIAM Conference on Parallel Processing for Scientific Computing (SIAM PP 2014), Feb 2014, Portland, Oregon, United States. ⟨hal-00978602⟩

Exporter

BibTeX XML-TEI Dublin Core DC Terms EndNote DataCite

Collections

UNIV-RENNES1 CNRS INRIA IRISA LABRI INRIA2 GENCI UR1-MATH-STIC UR1-UFR-ISTIC UNIV-RENNES UR1-MATH-NUM

298 Consultations

233 Téléchargements