LAL, LPT (Orsay), Labri/Inria Runtime (Bordeaux), CRI/école des Mines, Fontainebleau, LEREMA/Observatoire de Paris, etc.
Abstract
Iterative algorithms in computational field theory in
the petaflop heterogenous era
Context
projet ANR
Objectives
With the arrival of new types of supercomputers, namely the many-core
and multi-core computers, many computational sciences face new
challenges. While former may promise the top-performance for the
Linpack tests, many real-world applications are not able to take
advantage of the additional cores and/or accelerators. Recent trends
in computer science have indicated that it is imperative not only to
tune algorithms but often to change the computational strategy
completely to adopt the code to the petaflops-scale computers. Such
development requires close collaboration between computer scientists
and physicists to provide rapid improvement in scientific codes to
provide an instant gain top level machines.
Work program
We propose such interdisciplinary research to aid the computational
field theory, quantum chromodynamics, to adapt itself quickly to the
given architecture . This science is indispensable for interpretation
of the scientific results from the major scientific experiments, such
as Large Hadron Collider and SuperB Factory. Successful candidate will
participate in the research and development with a mixed team of
computer scientists and physicists, centered around France's main
computational centers - TGCC, CINES and IDRIS. These centers provide
access to a full variety of current architectures, and more is
anticipated in the future. The research will be centered around
possibility of automatic communication avoidance for the iterative
inversion algorithms, including domain decomposition. It will also
venture into development of communication layer of our new network
interface NEPAL, conceived for the amelioration of the communication
between accelerators.
Extra information
Prerequisite
Knowledge of English is required, as well as understanding of
parallel programming. C++ and CUDA/OpenCL may be a plus,
as well as openness to Physics Research challenges.
Détails
Expected funding
Research contract
Status of funding
Expected
Candidates
Utilisateur
christine.eisenbeis
Créé
Mardi 05 juin 2012 11:16:16 CEST
dernière modif.
Mardi 05 juin 2012 11:16:16 CEST
Fichiers joints
filename
créé
hits
filesize
Aucun fichier joint à cette fiche
Connexion
Ecole Doctorale Informatique Paris-Sud
Directrice
Nicole Bidoit Assistante
Stéphanie Druetta Conseiller aux thèses
Dominique Gouyou-Beauchamps
ED 427 - Université Paris-Sud
UFR Sciences Orsay
Bat 650 - aile nord - 417
Tel : 01 69 15 63 19
Fax : 01 69 15 63 87
courriel: ed-info à lri.fr