High-Performance and High-Productivity. Thomas C. Schulthess

Size: px

Start display at page:

Download "High-Performance and High-Productivity. Thomas C. Schulthess"

Bartholomew Hall
8 years ago
Views:

1 High-Performance and High-Productivity Computing ( ) Platform Thomas C. Schulthess

2 Four elements of the Swiss N Initiative Swiss Platform for HP2C ( ): Simulation capabilities that make effective use of next generation supercomputers Establish in CSE programs at Swiss universities Hardware Phase I ( ): Upgrade Cray XT system at CSCS to maximum possible within current infrastructure Develop new building infrastructure by 2012: State of the art infrastructure that support a machine footprint that is about a factor 10 larger than today Hardware Phase II ( ): Goal for CSCS is to host systems with performance of 20-25% compared to largest leadership system in the world

3 The ecosystem in numbers Spring (planned) ORNL/LCF 200 XT5 cabinets 1.5 PFlop/s Leadership 20 PFlop/s Sequoia LLNL LCF3 ORNL LCF3 ANL Blue NCSA 5x 5x CSCS only 20 XT5 cabinets => TFlop/s (infrastructure limited) 300 TFlop/s Regional/national center 4 PFlop/s Will require new building infrastructure at CSCS 5x 5x EPFL: ~50 TFlop/s UZH: ~60 TFlop/s ETHZ: ~70 TFlop/s 60 TFlop/s Local/institutional supercomputer 800 TFlop/s Think about this now!

4 High-risk & high-impact projects of the ( New procurement Cray XT processors Upgrade Cray XT proc. Dual core upgrade Cray XT cores 2008 Upgrade Cray XT cores 2009 Hex-core upgrade cores Final upgrade Cray XT Procurement next generation supercomputer N initiative Begin construction of new building New building complete

5 Overarching goal of Prepare computational sciences to make effective use of next generation supercomputers Specific goal Emerge with several high-impact scientific applications that scale and run efficiently on leadership computing platforms in 2012/13 timeframe Build on, and multiply the early science applications experience on Jaguar at ORNL DCA++: simulate models of high-temperature superconductivity first sustained petaflop/s in production runs (Gordon Bell Prize 2008) WL-LSMS: simulate termodynamics properties in magnetic nanoparticles sustained petaflop/s in production runs (Gordon Bell Prize 2009)

6 Elements of the DCA++ project Start with a real and challenging scientific problem Simulations to understand high-temperature superconductivity Take the best methods known in the field Quantum cluster theory and quantum Monte Carlo (DCA/QMC) Understand their limitations on today s emerging hardware QMC kernels are memory bandwidth limited Algorithmic reengineering to better map method to hardware Introduced delayed updates ( Ed updates) into QMC algorithm, method is no longer limited by memory bandwidth Take aggressive view on rewriting codes don t hesitate to rewrite major portions or even starting from scratch DCA++ was a total re-write, more general, and extensible Pay close attention to new hardware developments GPGPU work started in 2007 and motivated mixed precision solvers

7 Learning from the Oak Ridge experience: Covering all aspect of the simulation system Simulations Models, s, & Implementation Map to Hardware System operation System design Distributing DCA++ team: the (first tasks in sustained Switzerland: petaflop/s) T. A. Maier P. R. Domain C. Kent projects T. C. Schulthess G. Alvarez M. S. Summers E. F. D Azevedo J. S. Meredith M. Eisenbach D. traditional E. Maxwell role J. M. Larkin J. Levesque at Swiss univers. Sci. comp. research CSCS & U. of Lugano CSCS s ( Centers) Systems research D-INFK & vendors Example based on ORNL s early science teams that run on the first petaflop/s systems Physics (chemistry,...) Application mathematics Computer Science Computer Center Hardware vendor

8 High-Performance and High-Productivity Computing (HP2C) Domain areas 12 projects each with 2-3 FTE/yr. ~10 FTE/yr. EPFL ETHZ USI U. Basel Supported by HP2C platform Other networks (e.g. SNF/NCCR) Technology Platform CSCS ~10 FTE/yr.

9 Organization Advisory Committee Tom Dunning (NCSA) Thomas Lippert (Jülich) Phil Colella (NERSC) David Keyes (KAUST/Columbia) Bill Dally (NVIDIA) HP2C steering committee President of USI: Martinoli Rector of the University of Zurich: Fischer Rector of the University of Geneva: Vassalli President of ETHZ: Eichler President of EPFL: Aebischer Project Leadership Director CSCS: Schulthess (PI) USI Inst. f. Science: Krause (co-pi) EPFL Scientific Computing: Quarteroni (co-pi) Project coordination USI research services: Lepori, Colferai CSCS: Michele De Lorenzi (website) Core group: Scientific Research Group CSCS Thomas Schulthess (Lead); Sadaf Alam (lead future systems) Will Swayer (lead application/libraries) Jean Favre (lead visualization) John Biddiscombe Mario Valle Ugo Varetto Adrian Tineo (Jerome Soumage) Inst. f. Computational Science USI Rolf Krause + group Illia Horenko + group Projects: 1.Gyrokinetic simulations of turbulence in fusion plasma Villard EPF Lausanne 2.Ab initio molecular dynamics (CP2K) Hutter, U. Zurich 3.Computational Cosmology Lake, U. Zurich 4.Selectome: Darwinian Evolution Robinson-Rechavi, U. Lausanne 5.Cardiovascular Systems Simulations Quarteroni, EPF Lausanne 6.Algorithms for Quantum Interacting Systems Giamarchi, U. Geneva 7.High-Resolution 3D-Seismic Imaging Olaf Schenk, U. Basel 8.3D Models of Stellar Explosion Matthias Liebendörfer, U. Basel 9. Large-scale electronic structure calculations (BigDFT) Gödecker, U. Basel

10 Tier 1 projects of the platform Advanced Gyrokinetic Numerical Simulations of Turbulence in Fusion Plasmas Laurent Villard, EPF Lausanne New Frontiers in ab initio Molecular Dynamics Jürg Hutter (CP2K), Univ. of Zurich Computational Cosmology on the Petascale Geoge Lake, Univ. of Zurich Selectome, looking for Darwinian evolution in the tree of life Marc Robinson-Rechavi, Univ. of Lausanne for Cardiovascular Systems Simulations Alfio Quarteroni, EPF Lausanne Modern Algorithms for Quantum Interacting Systems Thierry Giamarchi, Univ. of Geneva Large-Scale Parallel Nonlinear Optimization for High Resolution 3D- Seismic Imaging Olaf Schenk, Univ. of Basel Productive 3D Models of Stellar Explosions Matthias Liebendörfer, Univ. of Basel Large Scale Electronic Structure Calculations (BigDFT) Stefan Gödecker, Univ. of Basel

11 Crosscutting thrust areas Visualization and data analysis for in-situ visualization and data analysis / revisit workflows Jean Favre, John Biddiscombe, (Jerome Soumage) initial focus on individual projects, but develop coherent strategy Performance analysis of (current) implementations routine job at CSCS, Cray Center of Excellence, IBM ZRL will help us establish metrics Algorithmic motifs / Phil Colella s dwarfs common algorithms but incompatible implementations establish/understand hierarchy of dwarfs common implementations at least a lower levels opportunity to add several additional students / postdocs at USI-ICS Will Sawyer + new hires@cscs with connections to Dongarra et al. Prototypes of future systems and programing environments Two areas we need to observe/engage: BG/Q & Hybrid Multi-Core PGAS languages (CAF, UPC, C++ solution under development) Programming environments for next generation node architectures

12 High-Performance and High-Productivity Computing (HP2C) Domain areas 12 projects each with 2-3 FTE/yr. ~10 FTE/yr. EPFL ETHZ USI U. Basel Supported by HP2C platform Technology Platform CSCS ~10 FTE/yr.

13 Computer systems - research and national supercomputer service User Projects Projects Prototype systems Advanced development systems National supercomputing production systems

14 QUESTIONS / COMMENTS?

Das Unsichtbare sichtbar machen wenn Supercomputer Prozesse simulieren. Thomas C. Schulthess

Das Unsichtbare sichtbar machen wenn Supercomputer Prozesse simulieren Thomas C. Schulthess Optimized winglets reduce environmental impact of aircraft Computational simulation of vortex formation in wake