Position details

The European Centre for Research and Advanced Training in Scientific Computing (CERFACS) is a research organization with seven shareholders (CNES, EDF, Météo-France, EADS, ONERA, SNECMA and TOTAL) that aims to develop advanced methods for the numerical simulation and the algorithmic solution of large scientific and technological problems of interest for research as well as industry, and that requires access to the most powerful computers presently available. Approximately 110 people, including more than 90 researchers and engineers coming from 10 different countries work at CERFACS in different research areas: parallel algorithms, code coupling, aerodynamics, gas turbines, combustion, climate, environmental impact, data assimilation, and electromagnetism and acoustics. 

Coupling numerical codes, i.e. implementing synchronized exchanges and spatial interpolation of information between these codes, is a central issue in many research fields such as climate modelling, data assimilation, or computational fluid dynamics. The “Climate Modelling and Global Change” team at CERFACS into which the trainee will work specializes in developing the OASIS and Open-PALM coupling software tools. 

OASIS (see  http://oasis.enes.org/) is used to couple numerical codes modelling the different components of the Earth System (oceanic and atmospheric general circulation, sea-ice, land, atmospheric chemistry, etc.) and by about 35 climate modelling groups in France and Europe but also in the USA, Canada, Japan and Australia. The last version, OASIS3-MCT is interfaced with the Model Coupling Toolkit (MCT) developed at the Los Alamos National Laboratory (USA) and allows parallel exchange and interpolation of coupling data. 

CERFACS, in collaboration with ONERA, is also developing the Open-PALM coupler (see http://www.cerfacs.fr/globc/PALM_WEB/index.html), which is currently used into many different research and industrial projects, ranging from operational data assimilation to multi-physics modelling, for example for studies in aerothermal predictions of aeroengines or fluid-structure interactions. Since the inclusion in 2011 of the ONERA CWIPI library, Open-PALM also offers parallel exchange and interpolation of coupling data.

Many coupled numerical applications are faced today to the need to efficiently run on the new many-core parallel computers that will offer exaflop performances before the end of the decade. This is the context into which CERFACS develops today the OASIS3-MCT and Open-PALM couplers.  

The objective of the training period is to compare the parallel performances and the scalability of the OASIS3-MT and Open-PALM couplers. In a first step, the trainee will start running a “toy” coupled application, i.e. coupling numerical codes that do not perform any real computing but that exchange realistic coupling fields, that can use either OASIS3-MCT or Open-PALM. This toy model reproduces the coupling between the numerical ocean general circulation model NEMO from LOCEAN (see http://www.nemo-ocean.eu/) running on a 0.25 resolution structured grid and the numerical atmospheric general circulation model is ARPEGE from Météo-France (see http://www.cnrm.meteo.fr/gmgec/arpege-climat/ARPCLI-V5.1/index.html) using a Gaussian Reduced grid of ~50 km. The trainee will then run this toy coupled application on at least 2 massively parallel computing platforms ranked in the first 50 of the TOP500 list (http://www.top500.org/list/2012/06/100), which CERFACS has access to. Many tests will be realized increasing the number of cores used and thereby quantifying the scalability of each coupler. If the performances do not increase above a certain number of cores, this behaviour will have to be analysed. If time permits, the trainee will adapt the toy coupled application to unstructured grids, for example icosahedral grids, and will repeat the scalability analysis of the couplers for this new type of grids more recently used in climate models.

The implementation work will be mainly done in Fortran 90 but the coupler themselves are coded in Fortran 90, C and C++. The trainee will be supervised by one research engineer, Sophie Valcke, and one researcher, Florent Duchaine.