Supercomputing at the University of Arkansas

1
Supercomputingat the University of Arkansas

• Amy Apon, Ph.D.
• Oklahoma Supercomputing Symposium
• October 5, 2005

2
Outline of Talk

• What is the status of supercomputing at the
  University of Arkansas?
• Also in relation to other institutions
• Why do supercomputing at our institution?
• How did we get this far?
• Acquiring Red Diamond
• What comes next?

3
The Status of Supercomputing at the University
of Arkansas

• Red Diamond supercomputer
• Number 379 on the Top 500 list, June, 2005
• 128 node (256 processor)
• 1.349 TFlops (trillion floating point
  operations/sec)
• First supercomputer in Arkansas
• 213K from NSF MRI grant, 08/04, Apon PI
• Co-PIs Pulay, Fu, Bellaiche, Deaton, Selvam,
  Mattioli, Thompsons, Johnston
• Substantial match from the University
• Substantial gift from Dell

4
Significance of Red Diamond

• Places the University of Arkansas among about 40
  peer academic institutions, public and private,
  holding this quality of resource
• As measured by the Top 500 list of the fastest
  computers in the world, released every June and
  November

5
Other Academic Supercomputing Sites (partial
list only the map shows the Members of the
Coalition for Academic Scientific Computation)
also include sites in Kansas, Nebraska,
Missouri, Alabama, South Carolina, Minnesota,
Wisconsin, Maryland, Delaware, and Oregon
6

• Ranking in Top 500 list, academics in the U.S. by
  state
• (Source http//www.top500.org)
• State Jun '05 Rank (29) Nov 04 Rank (34) Jun 04
  Rank (39) Nov 03 (41)
• Virginia 14 7 3
• Illinois 20,38,47,48 10, 22 5, 15, 406 4, 35,
  195, 259
• Penn 33,68,395 34, 222, 480, 484 25, 295,
  297 12, 142, 154, 156
• California 37,43,63,66,71,108,162 25, 31, 37, 62,
  444 23, 43, 44, 276, 387 63, 137, 171
• Utah 53 367 141, 230, 313 177, 248
• Oklahoma 54 253
• Mass 59 473
• Texas 74,427,441 40, 248, 449 58, 198, 270,
  411 26, 198, 260, 355
• New York 117,200,242,326 123, 152, 194, 308 95,
  147 38, 68
• Tennessee 129 413 199
• Alaska 136,342 76, 203 56, 154 72, 102, 355
• Louisiana 147 82 65 30
• Maryland 166
• Arizona 249 159 118 53

7
Why do supercomputing?An opportunity for funding!
Continuing supercomputing capability and federal
funding levels are correlated!!
8
Federal Funding Directions

• President's Information Technology Advisory Panel
  encourages the growth of "computational science,"
  or the use of computers to complement experiments
  and theoretical research.
• The panel calls for more federal spending on
  supercomputing (Source Chronicle Daily News
  04-15-2005)

9
Benefits to Campus Users of a Supercomputing
Center

• Over time, we can refocus existing resources to a
  high-quality centrally-managed facility avoids
  duplication of resources on campus
• Eliminate need for departmental and research
  group clusters
• Reduce cost for software licenses, startup funds
• Be a focus for supercomputing activity on campus
• Can be an attraction in recruiting top faculty
  and Ph.D. students
• We have infrastructure to support a larger system
• We can work on larger problems
• We become more competitive in grant applications

10
How did we get this far?Acquiring Red Diamond

• MRI Major Research Instrumentation grant from
  the National Science Foundation
• Only three MRI proposals can be submitted from an
  institution
• The first year we tried we did not make the
  campus cut
• Amazingly, not everyone believes that we need
  supercomputing!!

11
NSF MRI Proposal

• Funding is granted based on the quality of
  research
• Geeky computer science types need not apply
• Just evaluating the benefits of a
  high-performance network, multi-core processors,
  even compiler optimizations is probably not
  enough
• Need to demonstrate the need for computing power
  for science and engineering research

12
Computational Research at the University of
Arkansas

• Development of middleware tools (Array Files) for
  managing, locating, and indexing data for
  large-scale out-of-core computational chemistry
  applications.
• This research is inherently interdisciplinary and
  results are applicable to many other projects in
  this proposal.

13
Computational Research at the University of
Arkansas

• Computational chemistry in two major areas,
  including
• The development of a parallel Coupled-Cluster
  Singles and Doubles (CC-SD) code which will run
  efficiently on a distributed memory system, and
  
• The development of an efficient parallel version
  of our Fourier Transform Coulomb (FTC) method for
  large-scale density functional calculations.
• New formulas for new drugs!

14
Computational Research at the University of
Arkansas

• Materials science, using a state-of-art
  first-principles density-functional theory
  (DFT) computational approach.
• The research includes the study of novel
  nanostructure materials that possess unusual
  properties of technological importance, in
  particular
• Nanostructures of ferroelectric (FE) and
  piezoelectric oxides which exhibit many
  electrical, mechanical, and structural
  properties that are not shared by other materials
• Semiconductor nanomaterials!

15
Computational Research at the University of
Arkansas

• DNA sequence design and analysis of large sets
  of sequences for biotechnology and
  nanotechnology applications.
• The computing equipment proposed here will
  accelerate the search for large sets of
  non-crosshybridizing DNA sequences.
• These sequences will form a large library for use
  in DNA computations or nanotechnology
• New ways to store huge amounts of information!

16
Computational Research at the University of
Arkansas

• Multiscale modeling, including
• The computation of electronic and optical
  properties of nanodevices, the investigation of
  the issues in multiscale modeling
• Multiscale modeling of crack propagation in
  alloys and metals
• Models of tornados!

17
Computational Research at the University of
Arkansas

• Other projects
• Models of volcanos
• Next generation networking
• Geospatial databases
• Data mining

18
Observations from an MRI panel

• Base the equipment request on research drivers
• Request an appropriately sized resource
• For the problem
• And with appropriate subcomponents
• An error in the resource description is more
  easily forgiven than perceived deficiencies in
  the research, but either can kill the proposal

19
What Comes Next?UofA Current Challenges

• Education of researchers, faculty students
• Some of our best scientists still need education
  on how to use a distributed memory parallel
  computer, including MPI, compiler tools
• System administration
• Dont underestimate the amount of time to
  administrate a large system it does not scale
  linearly!!

20
What Comes Next?UofA Current Challenges

• Supercomputing operations
• Keep the AC on
• Power
• UPS
• Space
• Usage policies, administration
• How do you incorporate usage from new faculty?
• How do you partition usage fairly?

21
What Comes Next?UofA Current Challenges

• Future grant applications
• Lifespan of a supercomputer is about three years!
• Funding models for on-going operations
• How will basic systems administration and project
  director be funded?

22
What Comes Next?

• We are
• Increasing campus-level support for HPC
• Expanding our computational science and
  engineering activities
• New researchers and domain areas
• Collaborating (via grid computing)
• Within the state
• Regionally (OU, GPN, SURA)
• Expanding access to National Lambda Rail

23
Questions?

• Contact information
• http//hpc.uark.edu
• http//comp.uark.edu/aapon
• aapon_at_uark.edu