BubbleSched, plate-forme de conception d'ordonnanceurs de threads sur machines hiérarchiques

THIBAULT, Samuel; NAMYST, Raymond; WACRENIER, Pierre-André

doi:10.3166/TSI.27.345-371

Métadonnées

Licence d’utilisation du document

THIBAULT, Samuel
Laboratoire Bordelais de Recherche en Informatique [LaBRI]
Efficient runtime systems for parallel architectures [RUNTIME]

NAMYST, Raymond
Laboratoire Bordelais de Recherche en Informatique [LaBRI]
Efficient runtime systems for parallel architectures [RUNTIME]

WACRENIER, Pierre-André
Laboratoire Bordelais de Recherche en Informatique [LaBRI]
Efficient runtime systems for parallel architectures [RUNTIME]

Langue

Article de revue

Ce document a été publié dans

Revue des Sciences et Technologies de l'Information - Série TSI : Technique et Science Informatiques. 2008, vol. 27, n° 3-4/2008, p. 345-371

Lavoisier

Résumé en anglais

Exploiting full computational power of hierarchical multiprocessor machines with irregular multithreaded applications requires a very careful distribution of threads and data. To achieve most of the available performance, programmers often have to forget about portability and wire down ad hoc placement strategies that highly depend on the architecture. To guarantee the portability of performance, we have defined abstractions called ``bubbles'' for capturing both the hierarchical structure of the application's parallelism, and the hierarchical architecture of the targeted machine. We have defined a set of high level primitives to ease the implementation of dedicated, efficient and portable schedulers. We show the relevance of our approach and describe the mechanisms we developped for easily implementing such schedulers.< Réduire

Mots clés en anglais

Threads

Scheduling

Bubbles

NUMA

SMP

Multicore

Métadonnées

Partager cette publication !

Licence d’utilisation du document