Communication-aware scheduling on heterogeneous master-worker platforms

Jean-François Pineau 1, 2
2 GRAAL - Algorithms and Scheduling for Distributed Heterogeneous Platforms
Inria Grenoble - Rhône-Alpes, LIP - Laboratoire de l'Informatique du Parallélisme
Abstract : The results summarized in this document deal with the scheduling of independent tasks on large scale master-worker platforms, when realistic communication models are uti- lized. The contributions of this work are divided into three main parts: 1) Parallel algorithms: we underline the difficulty of scheduling identical independent tasks on heterogeneous master-worker platforms using the one-port communication model. We look at several sources of heterogeneity as well as several objective functions; 2) Matrix product: we compute the total communication volume that is needed for matrix multiplication in the presence of memory constraints and when data is centralized, and we develop a memory layout whose performance is close to the theoretical lower bound on the communication volume. We extend this algorithm for heterogeneous platforms; 3) Scheduling: lots of applications are constituted of a very large number of independent identical tasks. We focus on steady-state, and prove how to minimize the slowdown of one application when several are deployed, and how to minimize power consumption when only one application is present.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00530131
Contributor : Jean-François Pineau <>
Submitted on : Wednesday, October 27, 2010 - 3:14:03 PM
Last modification on : Friday, April 20, 2018 - 3:44:24 PM
Long-term archiving on : Friday, October 26, 2012 - 12:30:10 PM

File

Identifiers

  • HAL Id : tel-00530131, version 1

Citation

Jean-François Pineau. Communication-aware scheduling on heterogeneous master-worker platforms. Computer Science [cs]. Ecole normale supérieure de lyon - ENS LYON, 2008. English. ⟨tel-00530131⟩

Share

Metrics

Record views

317

Files downloads

209