Skip to Main content Skip to Navigation

Impact des modèles d'exécution pour l'ordonnancement en calcul parallèle

Abstract : In this thesis we study the behavior of parallel applications represented by a precedence graph. The programming of such applications is strongly relied to the execution support. We present and discuss the main execution models and their influence on the task scheduling problems. We study more deeply four scheduling problems on execution models that consider the communication costs. We propose one solution for a fine grain problem, the knapsack problem, on the hypercube on synchronous execution model in which the communication costs are implicit. We study the scheduling of independent chains on BSP. We show that the problem of finding an optimal scheduling is NP-hard. We also propose trade-off solutions on the number of communication/synchronization phases and on the processors idle time. The last two problems concern techniques to reduce the impact of the communication cost. The first one considers task duplication, we propose a list algorithm with performance guaranty 2 for the problems with small communication time on a limited number of processors. The second one consists of optimizing the communication phases by scheduling the messages transmission. As the search of the optimal solution is NP-hard, we propose several heuristics, and we analyze them experimentally.
Document type :
Complete list of metadata
Contributor : Thèses Imag <>
Submitted on : Wednesday, February 18, 2004 - 3:05:35 PM
Last modification on : Wednesday, May 19, 2021 - 3:37:40 PM
Long-term archiving on: : Wednesday, September 12, 2012 - 1:35:15 PM


  • HAL Id : tel-00004839, version 1



Alfredo Goldman. Impact des modèles d'exécution pour l'ordonnancement en calcul parallèle. Autre [cs.OH]. Institut National Polytechnique de Grenoble - INPG, 1999. Français. ⟨tel-00004839⟩



Record views


Files downloads