Sites Inria

Version française

CEA-EDF-Inria School

CEA-EDF-Inria School - Moving towards petaflop digital simulation on hybrid parallel architectures

This school is aimed at students, researchers or engineers interested in high-performance scientific computation, especially for the purposes of leading edge digital simulation in digital physics.

  • Date : 6/06/2011 to 10/06/2011
  • Place : Inria Sophia Antipolis-Méditerranée Research Centre
  • Organiser(s) : Loïc Gouarin, Stéphane Lanteri, Violaine Louvet, Jean Roman

Objective

Benefiting from the ever-growing performance of computer calculation resources, scientific computation now finds many industrial applications and applications of interest in terms of our society's quality of life (environment and health). To deal with the complexity of the problems arising these days in
research and engineering, parallel computing has become an essential step in making full use of these computing resources. Petaflop (and imminently exaflop) computation systems are emerging as non-uniform massively-parallel architectures. Such systems are based on non-uniform memory architecture (NUMA) nodes connected by rapid networks, each node being organised around a disparate set of processing units - multi-core processor boards, graphics processing units (GPU) or other types of accelerator boards.
These computing units are fed by complex multi-stage memory units, and taking full advantage of the computing power of the resulting systems (i.e. with sustained performance levels close to peak performance) requires software able to make effective use of this hierarchical organisation of memory units.
Within this context, the objective of the school is to introduce recent developments not only in programming models and environments, but also in numerical algorithms for supercomputing, targeting modern high-performance computing (HPC) architectures.

The school's schedule is based on the following topics:

  1.  Introduction to modern HPC multi-core based systems, GPU, FPGA, etc.
  2.  Low-level programming models and environments (in particular CUDA, OpenCL, HMPP) and their hybridisation with MPI or OpenMP ‐ Virtualisation - Communications in non-uniform systems
  3. High-level programming environments (such as PyCUDA, CLyther, etc.)
  4. Numerical algorithms adapted to modern HPC systems - numerical linear algebra for dense and sparse systems, algebraic solution algorithms (domain decomposition and hierarchical methods), FFT, etc.
  5.  Performance evaluation and optimisation methods linked to architectural features of HPC systems
  6. Open Source environments for scientific computation on modern HPC systems

 In addition, the school's schedule will also include presentations of practical experiments on the development of numerical methodologies for resolving problems stemming from computational physics.

Organisers

Luc Giraud

HiePACS project team, Inria Bordeaux ‐ Sud‐Ouest

Loïc Gouarin

LAGA, University of Paris 13

Stéphane Lanteri

Nachos project team, Inria Sophia Antipolis ‐ Méditerranée

Violaine Louvet

 Institut Camille Jordan, Lyon

Jean Roman

LABRi and HiePACS project team, Inria Bordeaux ‐ Sud‐Ouest

Key words

Scientific computation, high-performance computing, parallel processing, numerical linear algebra, numerical simulation

Programme

Monday 6 June 2011

8:45 - 9:00 a.m.

Introduction

9:00 - 10:30 a.m.

Session 1 - Part 1

10:30 ‐ 11:00 a.m.: coffee break

11:00 a.m. - 12:30 p.m.

Session 1 - Part 2

12:30 ‐ 2:00 p.m.: lunch

2:00 - 3:30 p.m.

Session 2 - Part 1

3:30 - 4:00 p.m.:amp;nbsp; coffee break

4:00 - 5:30 p.m.

Session 2 - Part 2

5:30 - 6:00 p.m.

Debriefing

Tuesday 7 June 2011

9:00 - 10:30 a.m.

Session 3 - Part 1

10:30 ‐ 11:00 a.m.:amp;nbsp; coffee break

11:00 a.m. - 12:30 p.m.

Session 3 - Part 2

12:30 - 2:00 p.m.: lunch

2:00 - 3:30 p.m.

Session 4 - Part 1

3:30 - 4:00 p.m.: coffee break

4:00 - 5:30 p.m.

Session 4 - Part 2

5:30 - 6:00 p.m.

Debriefing

Wednesday 8 June 2011

9:00 - 10:30 a.m.

Session 5 ‐ Part 1

10:30 ‐ 11:00 a.m.: coffee break

11:00 a.m. - 12:30 p.m.

Session 5 - Part 2

12:30 - 2:00 p.m.: lunch

2:00 - 3:30 p.m.

Session 6 - Part 1

3:30 - 4:00 p.m.: coffee break

4:00 - 5:30 p.m.

Session 6 - Part 2

5:30 - 6:00 p.m.

Debriefing

Thursday 9 June 2011

9:00 - 10:30 a.m.

Session 7 - Part 1

10:30 ‐ 11:00 a.m.: coffee break

11:00 a.m. - 12:30 p.m.

Session 7 - Part 2

12:30 - 2:00 p.m.: lunch

2:00 - 3:30 p.m.

Session 8 - Part 1

3:30 - 4:00 p.m.: coffee break

4:00 - 5:30 p.m.

Session 8 - Part 2

5:30 - 6:00 p.m.

Debriefing

Friday 10 June  2011

9:00 - 10:30 a.m.

Session 9 - Part 1

10:30 ‐ 11:00 a.m.: coffee break

11:00 a.m. - 12:30 p.m.

Session 9 - Part 2

12:30 - 2:30 p.m.

Final debriefing and buffet lunch

Keywords: Numerical simulation High Performance Computing Parallel computing Numerical linear algebra

Top