1 \select@language {french}
\r
2 \contentsline {part}{I\hspace {1em}Contexte scientifique et revue de l'\IeC {\'e}tat de l'art}{3}{part.1}
\r
3 \contentsline {chapter}{\numberline {1}Cadre de travail et contexte scientifique}{5}{chapter.1}
\r
4 \contentsline {section}{\numberline {1.1}Classe des algorithmes it\IeC {\'e}ratifs parall\IeC {\`e}les \IeC {\`a} large \IeC {\'e}chelle dans une grille de calcul}{5}{section.1.1}
\r
5 \contentsline {subsection}{\numberline {1.1.1}Partitionnement du probl\IeC {\`e}me}{6}{subsection.1.1.1}
\r
6 \contentsline {subsection}{\numberline {1.1.2}Modes d'ex\IeC {\'e}cution synchrone et asynchrone}{7}{subsection.1.1.2}
\r
7 \contentsline {section}{\numberline {1.2}M\IeC {\'e}thodes de r\IeC {\'e}solution parall\IeC {\`e}les du probl\IeC {\`e}me de Poisson et de l'algorithme two-stage multisplitting de Krylov}{9}{section.1.2}
\r
8 \contentsline {subsection}{\numberline {1.2.1}Algorithme de Jacobi}{9}{subsection.1.2.1}
\r
9 \contentsline {subsection}{\numberline {1.2.2}M\IeC {\'e}thode de r\IeC {\'e}solution GMRES}{9}{subsection.1.2.2}
\r
10 \contentsline {subsection}{\numberline {1.2.3}Solveur multisplitting}{9}{subsection.1.2.3}
\r
11 \contentsline {section}{\numberline {1.3}SIMGRID/SMPI : Simulateur d'ex\IeC {\'e}cution d'algorithmes parall\IeC {\`e}les MPI dans une grille de calcul}{10}{section.1.3}
\r
12 \contentsline {subsection}{\numberline {1.3.1}MPI - Message Passing Interface}{10}{subsection.1.3.1}
\r
13 \contentsline {subsection}{\numberline {1.3.2}Simulateur SIMGRID}{10}{subsection.1.3.2}
\r
14 \contentsline {section}{\numberline {1.4}Motivations}{10}{section.1.4}
\r
15 \contentsline {section}{\numberline {1.5}Conclusion partielle}{10}{section.1.5}
\r
16 \contentsline {chapter}{\numberline {2}Etat de l'art et travaux de recherche associ\IeC {\'e}s}{11}{chapter.2}
\r
17 \contentsline {section}{\numberline {2.1}Concepts et d\IeC {\'e}finitions}{11}{section.2.1}
\r
18 \contentsline {subsection}{\numberline {2.1.1}Performance de l'application parall\IeC {\`e}le et scalabilit\IeC {\'e}}{11}{subsection.2.1.1}
\r
19 \contentsline {subsection}{\numberline {2.1.2}Taux d'erreur lors de la pr\IeC {\'e}diction}{13}{subsection.2.1.2}
\r
20 \contentsline {subsection}{\numberline {2.1.3}Weak contre strong scaling}{13}{subsection.2.1.3}
\r
21 \contentsline {section}{\numberline {2.2}Probl\IeC {\'e}matique sur la pr\IeC {\'e}diction \IeC {\`a} large \IeC {\'e}chelle de la performance des applications}{14}{section.2.2}
\r
22 \contentsline {subsection}{\numberline {2.2.1}Facteurs li\IeC {\'e}s \IeC {\`a} l'\IeC {\'e}cosyst\IeC {\`e}me}{15}{subsection.2.2.1}
\r
23 \contentsline {subsubsection}{\numberline {2.2.1.1}Facteur architecture des processeurs}{16}{subsubsection.2.2.1.1}
\r
24 \contentsline {subsubsection}{\numberline {2.2.1.2}Facteur : M\IeC {\'e}moire et stockage}{18}{subsubsection.2.2.1.2}
\r
25 \contentsline {subsubsection}{\numberline {2.2.1.3}Facteur : R\IeC {\'e}seaux de communication}{22}{subsubsection.2.2.1.3}
\r
26 \contentsline {subsection}{\numberline {2.2.2}Facteurs li\IeC {\'e}s au code de l'application}{22}{subsection.2.2.2}
\r
27 \contentsline {subsubsection}{\numberline {2.2.2.1}Facteur : Taille du probl\IeC {\`e}me}{23}{subsubsection.2.2.2.1}
\r
28 \contentsline {subsubsection}{\numberline {2.2.2.2}Performance de la parall\IeC {\'e}lisation}{23}{subsubsection.2.2.2.2}
\r
29 \contentsline {section}{\numberline {2.3}Techniques de profiling et instrumentation des applications parall\IeC {\`e}les}{26}{section.2.3}
\r
30 \contentsline {section}{\numberline {2.4}M\IeC {\'e}thodes de pr\IeC {\'e}diction de la performance de l'application parall\IeC {\`e}le}{26}{section.2.4}
\r
31 \contentsline {section}{\numberline {2.5}Conclusion partielle}{26}{section.2.5}
\r
32 \contentsline {part}{II\hspace {1em}PARTIE II - Travaux de contributions, r\IeC {\'e}sultats et perspectives}{27}{part.2}
\r
33 \contentsline {chapter}{\numberline {3}Comparaison par simulation \IeC {\`a} large \IeC {\'e}chelle de la performance de deux algorithmes it\IeC {\'e}ratifs parall\IeC {\`e}les en mode asynchrone}{29}{chapter.3}
\r
34 \contentsline {section}{\numberline {3.1}Protocoles et exp\IeC {\'e}rimentations}{29}{section.3.1}
\r
35 \contentsline {section}{\numberline {3.2}R\IeC {\'e}sultats}{29}{section.3.2}
\r
36 \contentsline {section}{\numberline {3.3}Conclusion partielle}{29}{section.3.3}
\r
37 \contentsline {chapter}{\numberline {4}Simulation avec SIMGRID de l\textquoteright ex\IeC {\'e}cution des solveurs lin\IeC {\'e}aires en mode synchrone et asynchrone sur un environnement multi-coeurs simul\IeC {\'e}s}{31}{chapter.4}
\r
38 \contentsline {section}{\numberline {4.1}Protocoles et exp\IeC {\'e}rimentations}{31}{section.4.1}
\r
39 \contentsline {section}{\numberline {4.2}R\IeC {\'e}sultats}{31}{section.4.2}
\r
40 \contentsline {section}{\numberline {4.3}Conclusion partielle}{31}{section.4.3}
\r
41 \contentsline {chapter}{\numberline {5}Mod\IeC {\`e}le de pr\IeC {\'e}diction de la performance \IeC {\`a} large \IeC {\'e}chelle d'un algorithme it\IeC {\'e}ratif parall\IeC {\`e}le}{33}{chapter.5}
\r
42 \contentsline {section}{\numberline {5.1}Approche et m\IeC {\'e}thodologie}{33}{section.5.1}
\r
43 \contentsline {section}{\numberline {5.2}Exp\IeC {\'e}rimentations et r\IeC {\'e}sultats}{33}{section.5.2}
\r
44 \contentsline {section}{\numberline {5.3}Conclusion partielle}{33}{section.5.3}
\r
45 \contentsline {chapter}{\numberline {6}Conclusion g\IeC {\'e}n\IeC {\'e}rale et perspectives}{35}{chapter.6}
\r
46 \contentsline {section}{\numberline {6.1}Conclusion g\IeC {\'e}n\IeC {\'e}rale}{35}{section.6.1}
\r
47 \contentsline {section}{\numberline {6.2}Travaux futurs et perspectives}{35}{section.6.2}
\r
48 \contentsline {part}{III\hspace {1em}Annexes}{47}{part.3}
\r
49 \contentsline {chapter}{\numberline {A}Premier chapitre des annexes}{49}{appendix.A}
\r
50 \contentsline {chapter}{\numberline {B}Second chapitre des annexes}{51}{appendix.B}
\r
