More Related Content Similar to Jf lavignon (20) Jf lavignon2. Introduction
Sorry not to be present, hope to meet you soon
Agenda :
- HPC challenges as seen by a HPC supplier
- Cooperation example and idea for
Latin America – Europe cooperation
©Bull, 2011 2 Cooperation in extreme computing
3. HPC, strategic foundation for competitiveness, innovation and
defense
Defense – Geospace Performance &
Geo-Web Competitivity
HPC computing technology
Internet Imagery - Health
Multimedia & Virtual Arts & Sciences
Offering capabilities to design new products faster, to increase human and
environmental knowledge faster and to overcome security complexity
©Bull, 2011 3 Cooperation in extreme computing
4. Challenge (1/4) : concurrency and parallelism
History
- Gigaflops 1985 Cray 2 : 4 cores @ 1.5 ops/cycle
- Teraflops 1997 Asci Red : 4,500 cores @ 2 ops/cycle
- Petaflops 2008 Jaguar : 180,000 cores @ 4ops/cycle
- 10 petaflops 2011 K system : 700,000 cores @ 4ops/cycle
Next generation of systems
- Million to billion threads
- 16 ops/cycle for each threads
©Bull, 2011 4 Cooperation in extreme computing
5. Challenge (2/4) energy and power
Today general purpose computer
- 200 Mflops/W
- Petaflops 5MW
- 10 Petaflops in 2011 : 12,5 MW
Trend
- Use of accelerators (GPU, MIC) with better ratio flops/W than
standard processor but more difficult to program
- Use of low frequency standard processors
- Reduction of the amount of memory per core
Most the large centers do not plan more than 20MW for a
HPC system
©Bull, 2011 5 Cooperation in extreme computing
6. Challenge (3/4) data in HPC
Data inside the HPC system
- Memory access time
- Amount of memory
- Power cost of moving data
Storage of data (disk)
- Capacity still growing
- Bandwidth and seek time not progressing at the same pace
- Increasing parallelism
Data for scientists
- pre and post processing of huge volume of data
- Quality of data produced by large simulation (uncertainty
quantification)
- Semantic attached to data
©Bull, 2011 6 Cooperation in extreme computing
7. Challenge (4/4) : Resiliency
Next generation system
- Huge number of components : 10 to 10 memory chips
6 8
- Interface at high clock rate increasing bit error rates
MTBF will be low
- 10,000 components with a MTBF of > 10 years leads to a combined
MTBF of 10 hours
Robustness must be addressed at different levels
- System hardware
- System software
- Application
©Bull, 2011 7 Cooperation in extreme computing
8. Conclusion technological challenges
Important challenges for next generations of HPC systems
No one has the complete solution
Cooperation is a must
©Bull, 2011 8 Cooperation in extreme computing
9. Bull’s experience in R&D cooperation
Common laboratory
Technology partners Intel, Xyratex, Caps Entreprise…
One-to-one cooperation
Development of large infrastructures
Research institutions / campus
Project with local ecosystem
Projects at European level
International vision sharing
©Bull, 2011 9 Cooperation in extreme computing
10. Cooperation eco-system
Global for best of breed,
standardization
Customer cooperation for Technology
co-design and connection partners for
of market demands- complete and
supplier offerings innovative offering
Local for fertilization,
reactiveness and
teaming
©Bull, 2011 10 Cooperation in extreme computing
11. Example of instruments in Europe Systematic, Eureka, FP7
Systematic
- “World class” French Cluster, 142 enterprises, 311 SMB 92 research
centers & universities.
- Cooperation between SMEs, Large Companies, Research Centres and
National higher educational establishments
level - Supported by local authorities, economic development agencies, the
French Government and its partners.
- Leverage on Teratec for HPC
EUREKA
- 39 member countries plus the European Union.
European - The EUREKA label gives participants a competitive edge in their
dealings with financial, technical and commercial partners.
level
FP7
- EU's main instrument for funding research in Europe
- Grants are determined on the basis of calls for proposals
- Activities funded from FP7 must have a “European added value”.
©Bull, 2011 11 Cooperation in extreme computing
12. Systematic, FP7 & ITEA figures
Systematic FP7 ITEA
Eureka R&D
Type Cluster R&D program
program for IT
Time scale 2007-2013 2006-2012
Projects € 200 m / year 20,000 PY
€ 54 B
o/w fundings € 60 m / year € 600 m
o/w € 9 B for IT
Area France Europe Europe
©Bull, 2011 12 Cooperation in extreme computing
13. Example of a project : ParMa, Itea Gold Award 2010
Parallel Programming for Multi-core Architectures
A European project that gathers Facilitate the emergence of
users, software providers, HPC power-intensive innovative
system supplier applications by:
Objective: help the HPC - Developing advanced
community take full advantage technologies for exploiting
of multi-core architectures parallel computing
Deliver substantial performance - Exploring commonalities between
HPC, Multi-Core and MPSoC
improvements for: programming environments
- conventional HPC applications - Leveraging relevant technology
- mainstream applications and methodology between these
- embedded systems domains
Blast-furnace: €125,000
savings plus 16,000 tons of
CO2 emission/year
©Bull, 2011 13 Cooperation in extreme computing
14. To conclude
Latin America – Europe cooperation is
what I call “a global cooperation”
- Need to focus on global topics
- Ie programming models
- Scientific methods…
- Need to find mutual interest
- Economic system
- Educational system
- Need to involve “regional champions”
Cooperation at International level could
be boosted by new instruments
- Why not a "Eureka-like" program at
International level
©Bull, 2011 14 Cooperation in extreme computing