Ordonnancement efficace d'applications parallèles : les tâches malléables monotones

Abstract : The load balancing and data distribution are major problems to solve in order to implement a parallel application. They require to choose the date and location of the computations. The efficiency of the application is a function of these choices. We will solve this "scheduling problem" with a model recently proposed : the malleable tasks. The introduction to the domain of parallel computing includes the main drawbacks of some standard models. Namely, the fine grain modeling of application requires in these models accurate modeling of data exchange. The involved scheduling problems seem, in our opinion, difficult to handle. An application is handled by the malleable task model as a set of parallel tasks. Each one is executed simultaneously by several processors. The modeling of an application is the standard task graph but communications are taken into account implicitly in the execution time of each malleable tasks. We claim this approach simplifies the scheduling problem practically and theoretically. This document presents firstly the independent malleable tasks scheduling. We analyze previous works and propose a new algorithm in almost two shelves with a performance guarantee of 3/2. An average analysis of the algorithms is also presented. Some previous results for the problems with precedence constraints in related models are recalled. We propose a first approach to the problem of malleable tasks chains. Then, the ocean stream simulation is introduced. The practical use of the malleable tasks model to schedule this simulation is finally exposed.
Liste complète des métadonnées

Cited literature [107 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00006094
Contributor : Thèses Imag <>
Submitted on : Thursday, May 13, 2004 - 9:20:58 AM
Last modification on : Thursday, January 11, 2018 - 6:14:33 AM
Document(s) archivé(s) le : Friday, April 2, 2010 - 7:54:26 PM

Identifiers

  • HAL Id : tel-00006094, version 1

Collections

IMAG | UGA

Citation

Grégory Mounié. Ordonnancement efficace d'applications parallèles : les tâches malléables monotones. Réseaux et télécommunications [cs.NI]. Institut National Polytechnique de Grenoble - INPG, 2000. Français. ⟨tel-00006094⟩

Share

Metrics

Record views

377

Files downloads

1841