CERFACS CERFACS Scientific Activity Report Jan. 2010 – Dec. 2011

1 Introduction
The CFD (Computational Fluid Dynamics) team is the largest team of CERFACS. It gathers 60 to 70
researchers and more than half of them are PhD students. The team is organized around three main themes
and the present text is organized in the same way :
– combustion (leader : Bénédicte Cuenot),
– turbomachinery (leader : Nicolas Gourdain),
– aerodynamics (leader : Jean Francois Boussuge).
The CFD team develops codes in these fields for CERFACS shareholders and academic partners. To build
codes which are both efficient in terms of computer science and physics, the CFD team is not only
developing these codes but also applying them directly to multiple real-life CFD problems in aircraft,
engines, furnaces...
Important actions at CERFACS in 2012 are the consequences of the definition in 2011 of ’CERFACS
challenges’ for 2020. The two CFD challenges are COUGAR (the computation of a full gas turbine engine)
and PUMA (the computation of a full aircraft). After defining what the targets were, the team has prepared
these actions and many themes described below participate to this effort. At the same time, present actions
must be continued and the support for users of CERFACS codes on a daily basis must be improved because
more and more partners use CERFACS codes. This has also lead CERFACS to recognize that, in addition
to developing the codes themselves, preparing interfaces for users to utilize these codes efficiently is one of
the new missions of CERFACS.
The two CFD challenges correspond to multiphysic / multiscale projects. They will drive the CFD team
activity in the next years but also require significant collaborations with other CERFACS teams : typically
ALGO and PAE for the codes themselves and GLOBC for the coupler OpenPalm. Indeed the 2020
challenges can be met only by leading two simultaneous actions : (1) preparing CFD codes for exaflop
machines (with the ALGO team anda the HIEPACS INRIA/CERFACS laboratory) and (2) preparing
coupling methodologies to run multiple codes on a given exaflop machine (with the GLOBC team which
has a long-term expertise in the field of code coupling).
Massively parallel computing is and will still remain a central activity of the CFD team. Staying at the
forefront of HPC research in CFD is one of our goals. This is achieved by developing our own codes
for all new architectures but also by working at the European level to ensure that CFD needs are taken
into account in future projects. The CFD team is present for example in the EESI project (www.eesiproject.eu)
which prepares the European FP8 directions in the field of massively parallel computing and
in the scientific committe of PRACE (www.prace-project.eu) where scientists can have access to Tier0
machines. CERFACS is one of the few non american groups having access to the US super computers
through the INCITE project (www.alcf.anl.gov/collaborations/incite10.php). CERFACS also works with
machine vendors, for example Bull in 2011 to use the Curie Machine for AVBP or IBM for the BlueGene
platforms.
Inside CERFACS, AVBP remains one of the world most efficient Large Eddy Simulation codes on parallel
machines while elsA now allows aerodynamic simulations on thousands of processors. However, the CFD
team has started thinking about 2020 along a simple question : will these codes still be our workhorses in
ten years or should we start to work on a new code generation ? This work has lead to different initiatives :
one of them is the collaboration with CORIA (CNRS Rouen, Dr Vincent Moureau) to work for SAFRAN
on a new solver called YALES 2. Other paths are being discussed and this discussion is important because
it controls the team long term evolution.
CERFACS ACTIVITY REPORT 129