By continuing to browse the site, you are agreeing to our use of cookies. (More details)
Portal > Offres > Offre UMR5505-CHLBOU-027 - CDD chercheur (H/F) en Algebre Lineaire Computationnelle et Calcul à Haute Performance

Postdoctoral position for the European EoCoE-II project

This offer is available in the following languages:
Français - Anglais

Ensure that your candidate profile is correct before applying. Your profile information will be added to the details for each application. In order to increase your visibility on our Careers Portal and allow employers to see your candidate profile, you can upload your CV to our CV library in one click!

Faites connaître cette offre !

General information

Reference : UMR5505-CHLBOU-027
Workplace : TOULOUSE
Date of publication : Monday, November 25, 2019
Type of Contract : FTC Scientist
Contract Period : 18 months
Expected date of employment : 15 January 2020
Proportion of work : Full time
Remuneration : 2 617,05 euros gross monthly
Desired level of education : PhD
Experience required : 1 to 4 years

Missions

We are seeking candidates for a 18-months post-doctoral position at the IRIT laboratory of Toulouse. The subject, described below, is related to the improvement of methods for the solution of sparse linear systems on large scale parallel supercomputers. The successful candidate will work in the context of the European EoCoE-II project (\url{https://www.eocoe.eu/}) whose objective is to leverage the potential offered by the ever-growing computing infrastructure to Foster and accelerate the European transition to a reliable low carbon Energy supply using High Performance Computing.

Activities

Multigrid and multilevel methods are regarded as highly scalable
methods for solving systems of linear equations on large-scale
distributed-memory architectures [Adams et al. 2004; Falgout and Yang 2002]. They rely on two key components: an operator hierarchy
defining a cascade of levels, and efficient smoothers at each of these levels. Although for some sets of equations, such as symmetric
definite positive matrices, there are already some well established
methods and solvers [Jolivet et al. 2013], finding such optimal
components for an arbitrary problem is an open question. At scale,
these solvers typically build an increasing number of levels because
it is not always possible to define extremely aggressive coarsening
strategies. As such, and because a larger number of levels means
greater load inbalance, the computation to communication ratio can
deteriorate.
The successful candidate will have the opportunity to work on a variety of topics.

- Quantify the impact of the use of block low-rank factorizations
[P. R. Amestoy et al. 2019] as smoothers or coarse grid solvers. Indeed, a strategy to mitigate the increased volume of communication may be to stop building coarser and coarser levels. Usually, the coarsest level is small enough to be solved with a dense direct dense solver such as LAPACK. If, however, the operator hierarchy is less deep, the coarsest level will quickly become too large for this approach and it will be necessary to use a distributed direct sparse solver [P. Amestoy et al. 2001]. In such a case, the use of low-rank approximation techniques reduces the complexity of the coarse grid solution with a reliably controlled loss of accuracy. The mathematical properties of such a combination as well as its performance and scalability in a parallel setting will be studied.
- Propose new coarsening strategies to further decrease operator
complexities. Though this task is application-dependent, because it
mostly depends on the linear systems or the physics being considered, the candidate will have the liberty to choose between
the applications available through the EoCoE-II framework and focus
on one or multiple solvers.

Skills

- Strong expertise in sparse linear algebra
- Multigrid and/or direct sparse linear solvers
- High Performance computing parallel algorithms
- C(++) and/or Fortran programming, MPI and OpenMP

Work Context

The Institut de Recherche en Informatique de Toulouse (IRIT, https://www.irit.fr/) is one of the major French research centers in
computer science. The post-doc will be hosted in the Algorithmes Parallèles et Optimisation (APO, http://apo.enseeiht.fr) team located at the ENSEEIHT school of Toulouse.

We talk about it on Twitter!