PressCNRS international magazine

Table of contents

Information Technology

Computing Power

With the acquisition of a 207 teraflop supercomputer and the creation of the Grids Institute (Institut des grilles), CNRS is meeting a vital need of researchers from all disciplines, and moves to 3rd place in the world rankings of computational power.


© T. Goldmann/IDRIS-CNRS

In the foreground, ten Blue Gene/P units. Behind them is the IBM Power4 system, which will be replaced soon by an IBM Power6 system to complete the supercomputer.

French researchers are walking on air, as IDRIS 1, a CNRS center on the Orsay campus right next to Paris, has just acquired the most powerful operational computer in France. This supercomputer will provide a thirty-fold increase in the numerical computational power available to the French research infrastructure, reaching 207 teraflops 2 this September, compared to the 6.7 available to scientists today. This system, once fully operational, will boost CNRS to third position in the world rankings of supercomputing sites, just behind the Lawrence Livermore National Laboratory in the United States (500 teraflops), and the Jülich Research Center in Germany (soon to reach 240 teraflops). And that’s not one second too soon. “Over the last two years, we noticed that some researchers limited their requests for scientific computation–and this was of some concern to us,” said CNRS President Catherine Bréchignac, when the contract with IBM was announced on January 7th, 2008.
A growing number of disciplines, encompassing fields as diverse as nanotechnology and engine design, can no longer make do with small computers, or networked grids of computers. For example, researchers need massive computing power to run climate or ocean current models, study fluid mechanics, analyze protein structure for drug development, elucidate galaxy formation, and understand the ultimate components of matter.
As Victor Alessandrini points out, “our aim is to offer a computational service to the entire scientific community. The IDRIS facility can be considered a very large research instrument, just like a particle accelerator, a telescope, or an X-ray production facility.” Accessing the IDRIS facilities and the grey matter of its engineers is free, but follows certain procedures. “Every year, expert committees select four hundred projects based on their scientific merits and whether they tie in with our computational tools,” says Alessandrini. The CNRS, French universities, organizations such as the CEA (French Atomic Energy Commission) and even industry, within the framework of contracts with public research laboratories, all send in applications. Once a project is selected, IDRIS assists these research teams with 15 or so engineers and scientists responsible for operating and improving the machines. The Institute’s engineers and scientists, supported by industry experts, also train researchers from the public and private sectors in the most recent scientific computational techniques.

Denis Delbecq

The Grid Alternative
Grid computing is another way of providing users with massive data-processing capabilities, though supercomputers are still better for tasks requiring a high degree of parallelism.
Last December, CNRS launched the Grids Institute (Institut des grilles).1 This new institute coordinates all CNRS actions regarding computing grids and distributed computing systems. The objective is to consolidate existing grid production infrastructure, to reinforce research in this field, and to improve synergy between the various actors. Computing grids are a set of heterogeneous computing resources, geographically distributed and linked together via very fast networks. They provide grid users with the immense data-processing capacities required by many fields of science. “Particle physics is the number one user, currently taking up 80% of the French grid capacity,” explains Guy Wormser, director of the institute. That percentage could increase with the upcoming launch of the Large Hadron Collider (LHC)2 in 2008, expected to produce 15 petabytes3 of data per year. Life sciences, including fields as varied as genomics, proteomics, or medical imaging, are the second most important users of grids. Many other scientific fields may also be regular grid users. Collecting seismic data, mapping the universe’s black holes, or testing a bank of chemicals for specific molecular interactions all generate huge amounts of data that require large computer calculations. Grids themselves are also a subject of study for computer scientists, and the Grids Institute is an opportunity to bring together the experience of end-users and developers. Finally, as Wormser insists, “grids also favor international cooperation through the sharing of data, especially for southern countries with modest computer infrastructure, who are thus able to participate in major world projects.”

2. At CERN, on the border between France and Switzerland.
3. One petabyte = 1015 bytes

CONTACT: Guy Wormser,
Institut des grilles, Paris

Notes :

1. Institut du développement et des ressources en informatique scientifique.
2. A teraflop is a trillion calculations per second.

Contacts :

IDRIS, Orsay.
Victor Alessandrini,
Thierry Goldmann,


Back to homepageContactcredits