Logo du LIFL

Depuis le 1er janvier 2015 le LIFL et le LAGIS forment le laboratoire CRIStAL

  1. Doctoral studies

Thesis of

Pierre-Yves Aquilanti

Wednesday 14 December 2011
Amphi Turing, Bât M3 - LIFL

Repartited and Massively Parallel Krylov Iteratives methods applied to Geoscience on Heterogeneous Platforms Over the Petaflop

Directeur : Serge Petiton, Professeur, Université Lille 1
Rapporteurs : Pierre Manneback, Professeur, Université de Mons (Belgique)
Frédéric Magoulès, Professeur, Ecole Centrale Paris
Examinateurs : Jean-Luc Gaudiot, Professeur, University of California - Irvine (Etats-Unis)
Nahid Emad, Maître de Conférences, Université de Versailles
Henri Calandra, TOTAL SA
Nouredine Melab, Professeur, Université Lille 1
Guy Bergère, Université Lille 1

Krylov iteratives methods are numerical methods used to solve linear systems. They are used in various scientific domains because they are heavily parametrizable and modular. Moreover, they are relatively easy to parametrize which allow them to threat very important cases. However, the problem of these methods is that they are particularly difficult to masterize and optimize them require an heavy knowledge in numerical methods and computer science relatively to the application domain. 

We present differents parameters optimization technics as well as a distributed solving method in order to maximise the use of parallel homogeneous or heterogeneous hardware platform. Our optimization techniques aim to proceed to the automatic parametrization process of the method at runtime, thus allowing the user to avoid this part without lessening the processing efficiency. Our solving method presents the advantage of repartiting the computing modules and optimization algorithms in order to maximise the use of a computing platform. 

Our research results allowed us to enlight the interest of parameters autotuning techniques in parallel environments and their integration in existing software platforms. Our research shows the use of such technics for various computing and software platforms in relation with an industrial context.


UMR 8022 - Laboratoire d'Informatique Fondamentale de Lille - Copyright © 2012 Sophie TISON - Crédits & Mentions légales

Page respectant XHTML et CSS.

Pour tout commentaire / Comments and remarks : webmaster