July 5 th 2007 CC-IN 2 P 3

July 5 th, 2007 CC-IN 2 P 3 A high-performance computing facility for scientific research Visit of a Japanese delegation at CC-IN 2 P 3

CC-IN 2 P 3 presentation Main scientific organization in France CC-IN 2 P 3 federates the main computing resources National Institute for Nuclear Physics and Particle Physics Dedicated Computing Center For : • High energy physics • Nuclear physics • Astroparticle physics + some opening to other sciences 2007 Budget: 10 M€ + 2. 5 M€ salary D. Boutigny dapnia Atomic Energy Commission 2007 -07 -05 2

Mission Physics Experiments • Nuclear Physics • Particle Physics • Astroparticules Basic research Masses of data 101000 100111 00010010100011101 100010001111 00010 101000 100111 110001 111010 0001001010001101000 100111 00010010100011101 1110 10001111 00010 101000 00 11 000111 00010010100011101 10001111 00010 101000 100111 0001001010001 Data processing Publications D. Boutigny Physics Analysis 2007 -07 -05 3

Why such huge needs of computing for high energy physics ? n Scientific experiments are more and more sophisticated – Big number of electronic channels huge amount of data n International collaborations – Data is globally distributed – Data analysis is a geographically distributed process Development of high-performance wide-area network links n Simulation is an important part of the scientific process – Modeling of the detectors behavior is complex Computing and mass storage needs D. Boutigny 2007 -07 -05 4

Computing power evolution The compute farm purchased in 2006 entered the « Top 500 Supercomputer Sites » list 3700 jobs in simultaneous execution 12 000 jobs in queue throughput: 20 000 jobs/day D. Boutigny 2007 -07 -05 5

Evolution of the storage capacity 5. 4 PB TB Mass storage (tape-based) D. Boutigny Disk-based storage 2. 4 PB Ramp-up for the LHC era 2007 -07 -05 6

International Dimension Contributing to scientific research in an international context n Network of big centers for high energy physics data processing around the world – D. Boutigny n Data Active collaboration on several domains acquisition using instruments built by international collaborations 2007 -07 -05 7

The Large Hadron Collider (LHC): a new challenge ATLAS 4 experiments with a level of complexity never reached before CMS 10 billions collisions recorded each year 15 PB/year of data D. Boutigny 2007 -07 -05 8

Balloon (30 Km) 15 Peta. Bytes ? CD stack with 1 year LHC data (~ 20 Km) 1 year of LHC data 30 000 hard disks 50 CD-ROM Central processing of all those data is not viable = 35 GB It is necessary to integrate the data processing capabilities of several computing centers around the world D. Boutigny 6 cm 1 hard disk: 500 GB Concorde (15 Km) Mt. Blanc (4. 8 Km) 2007 -07 -05 9

A global grid infrastructure Annecy Nantes CC-IN 2 P 3 TRIUMF FZK ASCC CNAF Brookhaven T 0 NIKHEF NDGF T 3 (many) T 1 (11) RAL Lyon Marseille PIC Île de France T 2 (~100) Fermilab Clermont D. Boutigny Strasbourg 2007 -07 -05 10

EGEE Project Enabling Grids for E-scienc. E is a European project aiming at deploying a grid infrastructure for several scientific domains, coordinated by CERN Phase I ended – Phase II started since April 2006 • 90 partners, 32 countries • EU funding: 32 M€, 2 years Pilot New The European part of the LHC grid is operated by EGEE CC-IN 2 P 3 plays an important role in the EGEE gridwide operation 11 -12 people funded by EGEE Crucial for the success of the LHC grid in France D. Boutigny 2007 -07 -05 11

D. Boutigny 2007 -07 -05 12

Planned evolution of the site’s computing capacity CPU Disk The planned CPU and disk storage capacities grow faster than Moore’s law Huge impact on the cooling and power infrastructure 1. 5 M€ have been invested in the computer room upgrade 1 MW of electrical power for computing equipment A new building is foreseen for 2010 + 2. 5 MW D. Boutigny 2007 -07 -05 13

Grid usage HEP Grid / Total 29% Slope: +1% / month Main grid consumers: • LHC • Virgo Non HEP Grid / Total ~ 0. 4 % • ILC D. Boutigny 2007 -07 -05 14

LHC experiments CPU consumption 1 UI = 50 SI 2 k. h D. Boutigny 2007 -07 -05 15

Core competencies of CC-IN 2 P 3 (1) Operation of large computing farms Mass storage Usage of international connectivity through highperformance links • LHC: dedicated optical circuit CC-IN 2 P - CERN at 10 Gbps • Data distribution D. Boutigny Computing and data -intensive grid 2007 -07 -05 16

Core competencies of CC-IN 2 P 3 (2) High-performance storage Storage and retrieval of masses of data D. Boutigny 2007 -07 -05 17

Collaboration projects n ANR project IGTMD – Grid Interoperability and Massive Data Transfer – RENATER – LIP/RESO – CC-IN 2 P 3 + FNAL – Dedicated 2 x 1 Gb/s optical circuit between CC-IN 2 P 3 and FNAL n France-Japan Particle Physics Laboratory – NAREGI / EGEE interoperability – Software development on grid interoperability • New approach based on OGF standards: SAGA (Simple API for Grid Applications) / JSDL (Job Submission Description Language) – SRB / i. RODS developments and applications – 1 visitor from KEK will be working at CC-IN 2 P 3 for 8 months • Explore possibility to use i. RODS in LHC context France-China Particle Physics Laboratory n Future France-Korea International Laboratory n D. Boutigny 2007 -07 -05 18

Outreach n High energy physics – Traditionally high consumer of computing power n At CC-IN 2 P 3 – Development of a high quality infrastructure – Development of world-class core competencies n Usage of a fraction of the CC-IN 2 P 3 resources for other scientific or industrial domains Huge impact D. Boutigny 2007 -07 -05 19

Bio-medical applications at CC-IN 2 P 3 n Several groups use the CC-IN 2 P 3’s resources – – n Genomics Functional Medical Imaging Research on rheumatoid arthritis … Our contribution is focused on data management and usage of the compute farm – Helps on structuring and federating the user community n Opening possibilities for the research groups – One year worth of computing on a single (researcher’s) desktop performed at CC-IN 2 P 3 in some days or a few weeks D. Boutigny 2007 -07 -05 20

Collaboration, Communication, Dissemination Hosting of more than 200 web sites Palais de la Découverte (Paris science museum) for instance Operation of the visio-conference system shared by several nationwide scientific organizations (4500 h reservation in 2007) CERN – INSERM – CNRS - INRA Webcast: video recording and delivery of scientific events via the web Ø French Academy of Sciences Ø CNRS national and international conferences. . ~ 35 000 hits for Video On Demand in 2007 D. Boutigny 2007 -07 -05 21