220 likes | 225 Views
This presentation explores the importance of collaboration between research, education, and innovation in staying competitive in the knowledge economy. It discusses the European HPC mandate, challenges in HPC infrastructure investments, and ongoing projects and initiatives in the HPC ecosystem.
E N D
Ideas on High Performance Computingin Europe 44th HPC User Forum Stuttgart, 6 October 2011 Pekka KarpEuropean Commission - DG INFSO GEANT & e-Infrastructures “The views expressed in this presentation are those of the author and do not necessarily reflect the views of the European Commission”
research HPC education innovation The knowledge triangle at work The knowledge triangle at work To be a genuinely competitive in the knowledge economy, one must be better • in producing knowledge through research • in diffusing it through education • in applying it through innovation
Staying Competitive in Science Staying Competitive in Science • Collaboration between European and worldwide research teams; access to rare/remote resources • Global high-speed communication infrastructures • Global Virtual Research Communities • Data-intensive science and innovation • Use and manage exponentially growing sets of data • Experimentation in silico, simulation • Use of high-performance computing • ICT a fundamental enabler for research & innovation ••• 3
European HPC Mandate European HPC Mandate The Communication on ICT Infrastructures for e-Science Com(2009) 108 asks: • “Building a new generation of supercomputing facilities” • Member States to scale up and pool investment in support of PRACE • Commission to define and support an ambitious European strategic agenda for supercomputing Confirmed by the Conclusions of 2982nd Competitiveness Council of Dec. 2009
Challenges Challenges • Investments in HPC infrastructures require a long term perspective • Combine and reinforce the efforts of national and EU funding authorities - PRACE • Exploit the innovative potential of HPC services beyond science • Develop a new HPC strategy for industrial involvement • Transition to peta-scale and exa-scale computing creates new opportunities for both science and computing
Tier-0 –PRACE-1IP – €48 M Tier-1 DEISA2/PRACE-2IP – €53 M HP-SEE & LinkSCEEM2 EU National Tier-2 + EUDAT, EGI – €60 Mnumerous other projects Local HPC Eco-System HPC Eco-System Exa-scale prototypes - €42 M
CRESTA CRESTA • Collaborative Research into Exa-scale Systemware, Tools and Applications • 3-year IP Project (October 2011 - September 2014) • Total budget: €12 M€ (€8.6 M EC contribution), 1008 PMs • Building and exploring appropriate system-ware for exa-scale platforms • Enabling a set of key co-design applications for exascale • Employing both incremental and disruptive solutions • 6 full-scale science and engineering applications EPCC, HLRS, CSC, PDC, Cray, TUD, Allinea, ABO, JYU, UCL, ECMWF, ECP, DLR ••• 7
DEEP DEEP • Dynamical Exa-scale Entry Platform • 3-year IP Project (October 2011 - September 2014) • Total budget: €16.2 M (€8 M EC contribution), 1313 PMs • Implementation of a Booster based on MIC processors and EXTOLL interconnect • Energy-aware integration of components • Basic strategy to port applications: • Highly scalable kernels offloaded to the Booster part • Less scalable kernels executed on the Cluster part ••• 8
Mont-Blanc Mont-Blanc • To develop an European exa-scale approach based on embedded power-efficient technology • 3-year IP Project (October 2011 - September 2014) • Total budget: €14.5 M (€8.1 M EC contribution), 1095 PMs • Prototype HPC system based on European embedded processors • Design of a next-generation system • Open source system software stack • Up to 11 full-scale scientific applications ••• 9
PRACE – Results PRACE – Results • Partnership for Advanced Computing in Europe • 21 Countries joined forces to create a unique high-end High-Performance Computing Research Infrastructure • National co-investments of €400 M to deploy and operate up to four leading edge Tier-0 systems • First Tier-0 system (1 petaflops) in Germany (JUGENE) • High interest in PRACE access calls (6 times oversubscribed) • Address issues like more effective solar cells, biochemistry, fluid dynamics, particle and plasma physics, weather and climate models, material science, and astro-physics • Study "Development of a Supercomputing Strategy in Europe“ • PRACE AISBL opened on 5 October 2010 in Brussels by Commissioner Kroes • Second PRACE Tier-0 system (1.6 petaflops) in France (CURIE) • Third PRACE Tier-0 system (1 petaflops) being installed (HERMIT) • Study "Financing a Software Infrastructure for Highly Parallelised Codes“ 10
European HPC Service European HPC Service a.u. PRACE more than 15 PF PRACE 2 systems PRACE Tier-0 “DECI” DECI DEISA 2005 2006 2007 2008 2009 2010 2011 2013
10% Loss Study Findings Study Findings
Only half of USGDP spending Study Findings cont. I Study Findings cont. I
Top500 List Top500 List Asia 34% Europe 21% US 44%
European HPC Issues European HPC Issues • Europe has lost 10% of its HPC capabilities in the last 2 years while Asia and the US have increased their capabilities • Japan overtook Europe (all 27 Member States combined) in terms of HPC capacities available • Fragmentation of European HPC efforts across many countries • Some HPC production capabilities with reliance on foreign components and (sub) systems; European IPR benefitting others
Study Recommendations Study Recommendations
HPC: state of play HPC: state of play Science Industry HPC use HPC systemsupply Public service HPC Ecosystem EU market forhigh-end HPC: €630 M/yr Policy making 95% US Application software & tools
HPC: Europe’s place in a Global Race HPC: Europe’s place in a Global Race Basic premise:Europe should be a global HPC leader excelling in the application and production of HPC, in all domains(for industry, science and society) Alternative: A follower is just fine, what matters is the applications • Develop EU autonomous industrial capability • Alternative: continue to rely on systems from the US and others • HPC policy should be European • Alternative: Member States continue with their national (sub-critical) policies
Key Policy Actions Key Policy Actions • Develop EU-level governance • Spend more (MS, EU, industry, training) • Development of EU native capability through • Pre-commercial procurement • Level-playing field for EU supply industry • Increase HPC use in industry, especially by SMEs • Share application and software development with global partners
Priorities for HORIZON2020 Priorities for HORIZON2020 • Support international collaborations that are strategic for global scientific partnerships, thus reinforcing Global Virtual Research Communities • Consolidate e-Infrastructures as a multi-disciplinary platform for global collaborations • Reinforce European research capacity in the domain of high performance computing (HPC) • Adopt adequate organizational and governance models • Use e-Infrastructures as platforms for technology experimentation at large scale (e.g. Future Internet) ••• 20
Conclusions Conclusions • e-Infrastructures provide the underlying platforms for computationally intensive applications that enable international collaboration combining knowledge from different fields of science • HPC empowered e-Infrastructures integrate and make widely available national infrastructures and resources • e-Infrastructures implement a key EU policy and strategy -the European Research and Innovation Area • e-Infrastructures “facilitate” cohesion, standards, industry, etc. • New forms of organizations (Global Virtual Research Organisations) emerge relying on high performance computing ••• 21
The Bits in Science and Innovation ••• 22 ••• 22