Duration: 1 year
Period: to be started during 2015
Location: M2P2 laboratory-AMU computing center (MésoCentre), CNRS Aix-Marseille University, Centrale Marseille, France
Gross salary: from 2162 € to 2361 €/month depending on qualification and experience
Research project and job description
Title: Kinetic-fluid computing in magnetized fusion plasmas. Application to ITER.
Summary:
A one-year research engineer or postdoctoral position (depending on the experience) is opened between the M2P2 laboratory within the «Instabilities, Turbulence & Control» group and the Aix-Marseille University computing center. The objective of this job is to drastically improve the performances of our numerical codes developed in the frame of our activity on the numerical modelling of magnetized fusion plasmas. Numerical codes will have to run on massively parallel computers composed of several hundreds processors and using message passing, and OpenMP. This work will be carried out in close connection between applied mathematicians and numericists from Aix-Marseille University and plasma physicists at CEA Cadarache. It is part of the KFC project recently granted by the Aix-Marseille University foundation AMIDEX in the frame of the Initiative d’Excellence of the French government. The position is fixed-term up to 12 months with a start date first semester of 2014.
Job description:
The perspective of ITER, currently being built in Cadarache (south France), and within a shorter time frame, the setting up of an ITER-like full tungsten divertor in the Tore Supra tokamak at the CEA-Cadarache (WEST project, http://west.cea.fr/en/index.php), convey strong signals to intensify our effort on the numerical modelling of magnetized fusion plasmas. In tokamak reactors, the control of heat and particle fluxes on the divertor targets and the quality of plasma confinement in the reactor are two of the most crucial issues regarding on the performances. These call for an improvement of our capability to predict both turbulent and magnetohydrodynamic (MHD) transport, which leads to an important modelling effort. Benefiting from the standing increase of the available computing power, there exists a current international effort to develop efficient numerical codes able to describe self-consistently turbulent transport in realistic geometry. Due to the wide different scales involved, such simulations remain very challenging and require high performance computing (HPC).
The SOLEDG2D and TOKAM3X are currently being developed and used as part of this effort. Hybrid MPI+OpenMP parallelization is used to take advantage of the characteristics of large nodes machines currently available. Optimizing the performance of these codes will greatly improve the accessibility of complex cases allowing comparisons with tokamak relevant tokamak that require larger meshes and computing time. Optimization of the codes will be sought through both amendments of the codes themselves, but also the determination of good compilation flags. Both sequential and parallel (MPI and OpenMP) profiling will be needed. The expected output is the identification of bottlenecks and most promising sources of optimization. Special focus will be given to putting in place a good vectorization (AVX) strategy. The most time consuming parts of these algorithms could be also revisited like the linear sparse solvers. Some of the recommended modifications will be implemented in the code and tested. They will also be used to define a set of guidelines for future developments of the code.
Essential skills:
Candidates should have a demonstrated knowledge of HPC architectures, parallel programming and performance profiling techniques to achieve high performance on these systems, and associated software and tools. Advanced degree in Computer Science, Applied Mathematics or a related field required.
Contact person eric.serre<στο>univ-amu.fr