Scheduling independent tasks on multi-cores with GPU accelerators - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Article Dans Une Revue Concurrency and Computation: Practice and Experience Année : 2015

Scheduling independent tasks on multi-cores with GPU accelerators

Résumé

More and more computers use hybrid architectures combining multi-core processors and hardware accelerators like GPUs (Graphics Process-ing Units). We present in this paper a new method for scheduling efficiently parallel applications with m CPUs and k GPUs, where each task of the appli-cation can be processed either on a core (CPU) or on a GPU. The objective is to minimize the maximum completion time (makespan). The corresponding scheduling problem is NP-hard, we propose an efficient approximation algo-rithm which achieves an approximation ratio of 4 3 + 1 3k . We first detail and analyze the method, based on a dual approximation scheme, that uses dynamic programming to balance evenly the load between the heterogeneous resources. Then, we present a faster approximation algorithm for a special case of the previous problem, where all the tasks are accelerated when affected to GPU, with a performance guarantee of 3 2 for any number of GPUs. We run some simulations based on realistic benchmarks and compare the solutions obtained by a relaxed version of the generic method to the one provided by a classical scheduling algorithm (HEFT). Finally, we present an implementation of the 4/3-approximation and its relaxed version on a classical linear algebra kernel into the scheduler of the xKaapi runtime system.
Fichier principal
Vignette du fichier
CCPEhal.pdf (508.52 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01081625 , version 1 (10-11-2014)

Identifiants

Citer

Raphaël Bleuse, Safia Kedad-Sidhoum, Florence Monna, Grégory Mounié, Denis Trystram. Scheduling independent tasks on multi-cores with GPU accelerators. Concurrency and Computation: Practice and Experience, 2015, 27 (6), pp.1625-1638. ⟨10.1002/cpe.3359⟩. ⟨hal-01081625⟩
570 Consultations
584 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More