© 2007 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice Technology for better business outcomes NTIG Meeting Linköping, 1 April 2008 Petascale Computing and PRACE Martin Antony Walker BsC DIC PhD HPC Scientific Research Segment Manager HP EMEA
23 June 2015 Agenda The PRACE Project HP and Petascale Computing
33 June 2015 PRACE Partnership for Advanced Computing in Europe Preliminary phase study for the creation of a European research infrastructure under EC FP7 Included as “European High-Performance Computing Service” in ESFRI Roadmap for Research Infrastructures Report 2006 MoU signed by 16 countries in April 2007 PRACE project started Jan 2008 for 24 months
43 June 2015 PRACE Partners Principal Partners GENCI France GAUSS Germany NCF The Netherlands BSC-NSC Spain EPSRC UK General Partners GUP Austria CSC Finland GRNET Greece CINECA Italy UNINETT Sigma Norway PSNC Poland UC-LCA Portugal SNIC Sweden CSCS Switzerland
53 June 2015 PRACE Overview Vision −advanced computing in Europe for scientific leadership −world-class European HPC service for science & industrry −fund very large capability centers Mission −create three to five Tier 0 centers −build a persistent pan-European service −promote European presence and competitiveness in HPC including products and software Stakeholders: governments and funding agencies
63 June 2015 PRACE Work Packages I WP1: Management (management board chaired by Prof. Achim Bachem, FZJ Germany) WP2: Organizational Concept for Research Infrastructure: legal form, governance, funding, peer review, HPC ecosystem links, operating model WP3: Dissemination, Outreach, Training – includes industry and business collaboration WP4: Distributed System Management – will rely on system vendor solutions
73 June 2015 PRACE Work Packages II WP5: Deployment of Prototype Systems (3 to 4 – assessment will result in recommendations for production PFLOPS systems) WP6: Software Enabling for PFLOPS Systems – select applications, scale to PFLOPS. optimize, libraries and programming models – ˝the scientists will do it˝ WP7: PFLOPS Systems for – will include risk analysis of vendors WP8: Future PFLOPS Systems after 2010
83 June 2015 HP and Petascale Computing - PRACE Presented HP to WP7 in Paris 22 Feb 2008 (run by Francois Robin CEA/DAM) −report to management board in March 2008 −recommendations on prototype systems to management board May 2008 −decisions on prototypes June 2008 −recommendations on petascale machines to management board Nov 2008 −prototype deployments Dec 2008 HP is pursuing detailed discussions with principal and general partners
93 June 2015 HP and Petascale Computing – HPCD and HP Labs High Performance Computing Division being united with Scalable Datacenter Infrastructure team to increase investment and momentum HP Labs has been re-organized −Open Innovation Office to pursue and coordinate research collaborations −call for proposals to universities worldwide, inviting participation in joint research
103 June 2015 Prith Banerjee Sr. V.P. Research; Director, HP Labs Dean, College of Engineering, University of Illinois at Chicago Professor and Chairman, Electrical and Computer Engineering, Northwestern University; Director, Center for Parallel and Distributed Computing Director, Computational Science and Engineering Program, University of Illinois at Urbana- Champaign Research Interests: VLSI computer aided design; parallel computing; compilers
113 June 2015 Norman Jouppi HP Fellow Director, Exascale Computing Lab Fellow, ACM and IEEE Principal architect and designer, MIPS > 35 patents; > 100 technical papers Focus areas: −Emphasis on manageability, power, and availability −Parallel programming and software tools for the many- core era −Virtualization and integrated hardware/software platforms −Systems leveraging future technology (e.g., nanophotonics)
123 June 2015 Extending the vision: HP Labs Exascale Computing Lab Integrated “SmartCoolingInside” Tightly Coupled Power Management Room Level Infrastructures Global Scale “Clouds” Massive Scale-Out Storage Hetero- geneous Processin g Massive Scale Algorithms Fault Tolerant Self Healing Infrastructure Nano–Photonic Interconnects (On-die) Meso-Photonic Interconnects (i.e. Chip to Memory) Macro-Photonic Interconnects (beyond the board)
Thank you!