mcs/13-04-2004 HPC challenges in Switzerland Marie-Christine Sawley General Manager CSCS SOS8, Charleston April, 122003
Jan 15, 2016
mcs/13-04-2004
HPC challenges in Switzerland
Marie-Christine Sawley
General Manager CSCS
SOS8, Charleston April, 122003
mcs/13-04-2004
A short history of CSCS
1991- 1999
National service with a wide customer base
Since 2000
National service with strong leadership in computational science
2004-2010
National service with strong leadership for IT and scientific
computing support and development
mcs/13-04-2004
Outline of the presentation
• Brief recap on present status
• Strategy 2004-2007
• Major choices and challenges
mcs/13-04-2004
CSCS Autonomy
• Unit of ETHZ managed under performance
mandate and global budget
• New business models
mcs/13-04-2004
Driving forces
CapabilityAbility to serve the projects of the highest scientific quality requiring intensive resources
Sustainability
Clear line of funding, economy of scale and total cost of ownership
Scalability
Streamlining a set of services from the desktop to highest end servers
Flexibility Capacity to evolve according to the pace of technology
mcs/13-04-2004
New thrust areas for scientific computing
Discipline Description
Visualization Imaging, virtual reality,
Numerics & Benchmarking Optimisation, selection and development of solvers and
other mathematical methods
Data Intensive Computing Data mining, knowledge management, information
management and retrieval, machine learning,
geographical information systems
Grid and Distributed
Computing
Grid computing, networking, communications, agents,
distributed services
Visiting scientists program Organisation of courses and summer schools, visiting
scientists
Modelling Framework
Support
Hosting and fostering a software framework that is being
used as the community model of a research discipline
mcs/13-04-2004
CSCS in 2007
HPCN31%
Scientific computing
59%
Administration
8%
Expertise development at CSCS
CSCS Today
HPCN65%
Scientific computing
15%
Administration20%
mcs/13-04-2004
Usage by field, 2003
Bio/Chemistry9%
Engineering26%
Material Science
8%
Physics16%Mathematics
0%
Environment41%
PVP NEC SX-5
MPP IBM Power4
Bio/Chemistry51%
Environment9%
Material Science
32%
Mathematics0%
Physics7%
Engineering1%
mcs/13-04-2004
CSCS configuration
mcs/13-04-2004
HB networksmake locality less of an issue
ETHZ
•HP Superdome
•256 Opteron Cluster
EPFL
•HP SC 45, 100 Alpha
•Origin 3800 128 Mips
•100 P4 Cluster
UNIZH
•zBox 288 Intel IA32
•512 Opteron Cluster, myrinet
PSI
•64 Opteron Cluster
CSCS
•NEC SX5 16 Vector
•p690 IBM 256 Power4
mcs/13-04-2004
Meteo and clima modelling frameworkEach day, MeteoSwiss produces its numerical weather predictions, including daily operational short range weather forecasts and research activities are run at CSCS.
mcs/13-04-2004
Biological material modelling(prof. M. Parrinello)
mcs/13-04-2004
Fundamental CFD (prof. L. Kleiser)
mcs/13-04-2004
Particle physicsLHC in CH
mcs/13-04-2004
Roadmap
03 04 05 06 07
HEP
Bio
AstrophysReactor phy
SMP/vector Next generation
08
MPP
Visualization
Funding
mcs/13-04-2004
Risks and challenges
• Scientific group needs
– Facilitate multi disciplinary
collaboration
– Improve operating efficiency
– Tools for extracting scientific
information from complex and
massive data
• IT Needs
– Optimize Infrastructure,
computing on demand
– Integrate distributed resources
– Enable data access, integration
and curation
• Risks
– Dispersion of resources
– Not enough impact
– Balkanisation
• Weaknesses
– Insufficient coordination
– Resource allocation procedure
• Opportunities
– Establish a strong collaboration
platform
• Strengths
– Scientific portfolio
– Flexible organizations
mcs/13-04-2004
Please visit us
• On www.cscs.ch
• And for SOS9, March 2005