Computer Science in UNEDF

1
Computer Science in UNEDF

• George Fann, Oak Ridge National Laboratory
• Rusty Lusk, Argonne National Laboratory
• Jorge Moré, Argonne National Laboratory
• Esmond Ng, Lawrence Berkeley National Laboratory
• Ken Roche, Oak Ridge National Laboratory
• Masha Sosonkina, Ames Laboratory

2
Goals of the Computer Science Effortsin the
UNEDF SciDAC

• Work with specific physicists in the project to
  improve particular UNEDF component codes
• Create methods and software that are useful to
  other codes in UNEDF
• Create methods and software that may be useful to
  other high-performance scientific applications

3
The Collaborations

• Large-Scale Eigenvalue Calculations
• James Vary (Iowa State), Esmond Ng, Chao Yang
  (LBNL)
• Asynchronous Dynamic Load-Balancing Library
• Steven Pieper (ANL), Rusty Lusk (ANL)
• Multiresolution Methods for Nuclear DFT-Based
  Methods
• Witold Nazarewicz (UT), David Dean (ORNL/UT),
  George Fann (ORNL)
• Coupled Cluster Expansions
• David Dean (ORNL/UT), Ken Roche (ORNL)
• Performance Analysis and Optimization
• Witold Nazarewicz (UT), Jorge Moré, Boyana
  Norris, Jason Sarich (ANL)
• Implementation of 3D-lattice scalable nuclear
  static and time-dependent Density Functional
  Theory (DFT)
• Aurel Bulgac (UW), Ken Roche (ORNL)
• Usability and Automatic Deployment
• James Vary (Iowa State U.), Masha Sosonkina
  (Ames Lab)

4
Where the CS Collaborations Fit In
5
INCITE Award

• Title Computational Nuclear Structure
• Awardees (all in UNEDF)
• David Dean, Oak Ridge National Laboratory
• James Vary, Iowa State University
• Witold Nazarewicz, University of Tennessee and
  ORNL
• Steven Pieper, Argonne National Laboratory
• Computer Resources awarded
• 10,000,000 processor-hours on Argonnes IBM
  BlueGene P
• 7,500,000 processor-hours on Oak Ridges Cray
  XT-4
• Much of the computer science work described here
  is targeted at scaling up the relevant codes to
  take advantage of the power of the leadership
  class machines.

6
Testimonials

• We have made considerable progress and much more
  is in the pipeline. These collaborations between
  physicists and computer scientists have been
  critical to making rapid progress. -- James
  Vary, Iowa State
• I think all us are working as a team very well
  on this project which ultimately will enable the
  GFMC program to use 10,000's of processors.
  Without this collaboration, I see no way that the
  GFMC program would be able to effectively use the
  next generation of supercomputers. I also think
  that my needs are helping to define a package,
  being written by computer-science professionals,
  that will be useful to others. -- Steve Pieper,
  Argonne National Laboratory
• In general, there is a great level of
  satisfaction on the collaboration with our CS
  colleagues. -- Witold Nazarewicz, Oak Ridge

7
End of CS Overview