Inria and CERFACS create a joint laboratory to develop tools and methods dedicated to high-performance computing
© Inria / Photo Kaksonen
The National Institute for Research in Computer Science and Control (Inria) and the European Centre for Research and Advanced Training in Scientific Computation (CERFACS) today announce the creation of a joint research laboratory to respond to the challenges of high-performance digital simulation.
High-performance computing at the heart of major technological and scientific advances
Intensive scientific computation and digital simulation have become particularly important in most scientific fields and for the majority of technological applications. Digital simulation or in silico experimentation are often the only available means for analysing complex systems and solving problems with critical implications for the future that are impossible to test in any other way. With the aid of increasingly refined modelling techniques and increasingly rapid algorithmic tools executed on high-performance computers, researchers and engineers have obtained ever more relevant responses to the scientific challenges posed by medicine, biology, climatology and issues connected with energy and the environment. To support these developments and enable both scientists and manufacturers to access ever more powerful computers, research needs to move forward.
Ensuring "efficient scaling" or how to face the challenges posed by increasingly powerful computers
The arrival in 2011 of non-specialised computers with processing power in the region of petaflops (1015 floating-point operations per second), and shortly thereafter exaflops (1018 floating-point operations per second), calls into question the entire simulation chain. All of the computation components in this chain will have to "efficiently scale", in the sense that they must be able to execute effectively on a very large number of processor cores. This issue is being addressed by the international scientific community both in Europe (PRACE project) and the US (IESP project).
A joint laboratory that responds to the need for a multidisciplined approach
Research work on the development of tools and methods dedicated to high-performance simulation covers the full spectrum of skills, from mathematical modelling to the full-scale verification of multi-physical and multi-scale complex simulations that need to be executed on the most powerful parallel computers available at a given time. This work is, by nature, multidisciplined and concerns in particular applied mathematics, computer science and the application disciplines that play a key role in the field of simulation. This scientific skill set occupies a central position in CERFACS' and Inria's know-how.
The scientific activities of the Inria/CERFACS Joint Laboratory for Supercomputing will be structured as "Research Initiatives" involving personnel and resources from both organisations. The first initiative envisaged is a joint Project Team (HiePACS - High-End Parallel Algorithms for Challenging Numerical Simulations) between Inria Bordeaux – Sud-Ouest, the Research and Higher Education Cluster (PRES) of Bordeaux, the French National Centre for Scientific Research (CNRS - LaBRI UMR 5800, a laboratory associated with Inria) and featuring permanent researchers from the "ALGO" team at CERFACS. The director of this joint laboratory will be Jean Roman, Professor of Computer Science at the Polytechnic Institute of Bordeaux and currently on temporary assignment at Inria.
Chairman and CEO of Inria
« By uniting our two organisations' research and development potentials in the field of high-performance computing, we are convinced that we can have a larger scientific impact. The creation of this laboratory reflects our intention to contribute to the significant advances in this field. This initiative forms part of the same dynamic as the creation of our joint laboratory with the University of Illinois at Urbana Champaign in June. »
, Director of CERFACS.
« By creating this joint laboratory with Inria we are bringing together our complementary skills to develop essential algorithmic methods for getting the most out of tomorrow's supercomputers. »