Un «CERN» de l'environnement et du climat bientôt à Genève www.bilan.ch
Bob Bishop is audacious. The former CEO of Silicon Graphics (SGI), Australian by birth and Genevois at heart, wants to build a giant computer in Geneva. The machine he envisages will be able to execute a million billion operations per second (a ‘petaflop’), as a starting point. To comprehend thisamount of computing power, imagine each person on the planet performing 150,000 operations persecond!
“Yet it’s a minimum level,” calmly explains the manager, well-seasoned in the supercomputer
market. “When one wants to model the whole Earth and especially understand the interactions
between the oceans, the atmosphere and biodiversity, in order to foresee climate changes or to
anticipate natural catastrophes, a petaflop is the minimum unit of performance to consider”
Modeling the whole Earth on a computer naturally seems like an outlandish idea. Technically
however, it’s not unrealistic. With its 16,384 cores of parallel functioning, the new IBM Blue Gene/P
at the EPFL reaches a peak of 56,000 billion operations per second. In the United States, there are
four supercomputers which have passed the petaflop mark recently (two of which are publicly
visible). To attain this level, and even exceed it by a factor of a thousand (an exaflop), this
tremendous compute power would require a new generation of machines to appear. Such machines
most likely will be based on customised microprocessors like those developed by Tensilica (as in the
Green Flash Machine conceived in collaboration with the University of Berkeley), or on the new Fermi
GPU chips by NVIDIA, or the like.
“There are 100 billion neurons in the human brain,” continues Bob Bishop, who also presides over the
advisory board of the EPFL Blue Brain project which aims to reverse engineer the brain. “If one
divides the Earth’s surface into units of one kilometer cube in the spherical shell from 100 kilometers
above the Earth to 100 kilometers sub-surface, one also arrives at this figure of 100 billion units.”
And naturally, the possible interactions between these units are even more numerous - hence the
need for supercomputers in both projects.
The opportunity of large numbers
According to Professor Martin Beniston, leader of the Climatic Change and Climate Impacts Research
Group at the University of Geneva, “the best global climate models currently work with grids where
each point is at best 50 kilometers apart. If the definition becomes on the order of one kilometer,
our forecasting models will be considerably improved – even without taking into account that a
holistic approach integrating data from all the Earth sciences will be able to reveal the unsuspected
interactions between, for example, geology and climatology.”
“In the decades to come, the power of computer modeling and simulation will allow us to grasp
problems of this magnitude,” resumes Bob Bishop. “There is therefore a historic opportunity here.”
So as not to miss it, these past months he has drawn up a plan of action with a battery of experts
from universities and international organizations, such as the WMO in Geneva. A plan discovered
exclusively by Bilan.
In substance, the project envisages creating a center which will bring together some 200 computer
scientists, climate and geoscience specialists in Geneva, and a second building within a radius of 20
kilometers to host the supercomputer itself, possibly in France to take advantage of inexpensive
nuclear power from EDF. Over a period of eleven years, the investment in this International Centre
for Earth Simulation (ICES) would amount to 800 million Swiss francs (which is today approximately
USD 800 million).
As this represents a large sum of money, Bob Bishop doesn’t consider for a moment that this center
would be only for Swiss use. His project plans to connect climate and environment research centers
around the world with this center, in the same manner as CERN does in the realm of particle physics.
For months, he has been visiting many of the labs implicated, and has obtained the support of a good
portion of the scientific community.
“A full global simulator of the Earth’s total activity doesn’t exist today. Most of the centers which
simulate weather and climate work on a local or regional scale only.” In fact even the most visible
Earth Simulator - started in 2002 in Yokohama, Japan - is only planetary by name. It concentrates
mostly on Japan and its regional environment rather than on the whole Earth, for good reasons: the
complexity of natural phenomena which surround the Japanese archipelago (typhoons, Earthquakes
and tsunamis). And foreign researchers who want to test their models normally must actually go to
Yokohama to access the computing power of the Japanese center (which is presently 122 teraflops).
On the contrary, ICES will be an open international center, highly networked. It’s even the reason
why Geneva was chosen as the location to establish it. Bob Bishop foresees in effect that the project
will be instituted as a Swiss foundation and financed in the framework of a private-public partnership.
A third of the 90 million francs per year needed for the center’s operations would come from private
foundations. Another third would come from service contracts with specific companies such as
insurers, reinsurers, mining or energy groups. The last third would need to be financed with public
money (or ‘in kind’ funding) with the support of governments.
On this last point, one can of course ask oneself to what extent large countries would accept to leave
such a project simply up to Switzerland? Bob Bishop’s answer gives an idea of his diplomatic talents.
“As each country needs the data of all the countries, China or Russia would probably not want to
participate if the project is undertaken in a very dominant country like the United States, Japan or
Great Britain.”
The assets of Geneva
The stability and neutrality of Switzerland would therefore be one of the assets in this context. Bob
Bishop cites as proof the World Meteorological Organization, with which he collaborates. The latter
federates some 180 national meteorological institutions. And it is currently working on projects to
integrate local data collected by its members into a quality single standard and format.
One can also add the experience of CERN to the argument for Geneva, with its proven history of good
functioning in Switzerland. Bob Bishop adds, however, that it took fourteen years to put the LHC
(Large Hadron Collider) in place. “We hope to go faster.” His timetable plans a start in 2011. “It
won’t be too early to start providing a stronger scientific basis for major decisions on large scale
projects, especially when they interact strongly with our environment.”