n°28I quarterlyI January 2013 Focus | 25 w hIGhPErFormAnCE ComPuTInG 09 The US is still the undisputed dedicated to the Large leader with regard to High Hadron Collider (LHC) Performance Computing experiments and to èque (HPC). In 2012, the country biomedical and industrial H boasted 252 of the world’s applications. “More than Hotot 500 most powerful super- 2500 researchers and cnrsP . e akacs, computers, about half of the academics are involved ; D ar total computing power in high performance I DD rés . l o available worldwide— computing at CNRS, F g / I l estimated at 120 petaflops making us one of the © D o , (PFlops).1 With 22 of these largest multidisciplinary including HPC, computing 09 thecurie kIM . r 500 supercomputers communities in Europe,” grids, cloud computing, and supercomputer c ong MB I u representing a total of 6.4 points out Michel Daydé, large data centers,” notes can perform up - e / gI M D ar ert, Bac PFlops, France ranks 6th in director of the CNRS HPC Jean-Pierre Vilotte, scientific to 2 million billion DD B elaro, operations per o u g H / terms of computing power Orientation Committee director at CNRS’s INSU.3 The second, and store c nasa . , dedicated to HPC. (COCIN). But the number of COCIN plans to implement the equivalent of . ler, H r uarez, I At CNRS, the IN2P32 hours allocated either to this strategy both at the 7600 years of music D I n HI computing center plays national centers or European national and European level. files. It is located at s c . t enecker, Pawson, a key role in this field by projects does not always 01.apetaflopequalsonemillion bruyères-le-châtel I developing computing grids satisfy the community’s billionfloating-pointoperations in theessonne I rI sual . growing HPC needs. “In persecond. region (France). s u, vI lI order to meet the demand, 02.nucléaireetdephysiquedesphysiquedenationalInstitut © we must keep increasing particules. the overall capacity of 03.Institutnationaldessciencesde supercomputing power in l’univers. the country, and associate it contact InFormatIon: to a suitable organization michel Daydé (urbanization) of computing Jean-Pierrevilottemichel.dayde@cnrs-dir.fr> and data infrastructure > vilotte@ipgp.fr 04the simulation designed to federate distributed data and of the universe knowledge in biomedical imaging. developed for the The Credible project has an opposite Deus project will approach to that of ADNI. While ADNI generate more than 150 Pb of data. centralizes biomedical information be- 05studying the fore analyzing it, Credible federates the climate (here a study data scattered across a number of French of the atmospheric hospitals. “Sooner or later, the strategy humidity on June 17, 1993) requires of centralizing huge amounts of processing colossal medical data will be limited by available amounts of data, storage capacity,” explains Montagnat. which should reach “Furthermore, the proliferation of data one zettabyte (zb) by 2020. acquisition devices in hospitals makes it impossible to prevent the scattering of consortium founded in 1994 by Tim 06 Presentation field: biomedical research. In the US, the information across various storage sites.” Berners-Lee, principal inventor of the of the diseases Alzheimer’s Disease Neuroimaging But this federated approach is also prone World Wide Web. Some preliminary americans and their Initiative (ADNI) has become one of the to technical difficulties.affecting 7.2 million versions already exist for this new interrelationships largest public image databases dedicated Two types of data must be federated: so-called “semantic Web,” which makes provided by themIt. to a single pathology. It compiles raw data (images, treatment results, etc.), information on the Internet intelligible to 07an example Magnetic Resonance Imaging (MRI), and so-called “symbolic” data, which machines. Making the semantic Web a representation: nuclear medicine examinations, and describes this raw information (the con-of complex data universal model is a daunting task, how- here, a map of spinal taps for cerebrospinal fluid. ADNI text in which it was collected, the ana- ever. “It’s one thing to graft query algo- worldwide scientific already contains more than 5000 MRI tomical or pathological characteristics rithms onto a centralized database,” collaborations images from 2000 patients. extracted from an image, etc.). “One of between 2005 adds the LIG scientist, “but managing it and 2009. “With the advent of digital technolo- the Credible project’s main challenges is on the Internet—a gigantic decentralized gies, the amount of information from to develop a semantic representation of entity—is more than a question of tech- biomedical imaging has considerably this symbolic data to give it scientific nological improvement.” increased over the past decade,” says meaning,” he continues. Once harmo- Johan Montagnat, CNRS senior nized, these usually very heterogeneous wEll-DIsTrIBuTED DATA researcher at the I3S2 laboratory in databases can be used for a single national In recent years, new technologies drasti- Sophia Antipolis. He is also Mastodons or even international clinical study. cally changed practice in one scientific coordinator for the Credible project, Analysis of biomedical resources on that
CIM28
To see the actual publication please follow the link above