[House Hearing, 108 Congress]
[From the U.S. Government Publishing Office]
SUPERCOMPUTING: IS THE U.S.
ON THE RIGHT PATH?
=======================================================================
HEARING
BEFORE THE
COMMITTEE ON SCIENCE
HOUSE OF REPRESENTATIVES
ONE HUNDRED EIGHTH CONGRESS
FIRST SESSION
__________
JULY 16, 2003
__________
Serial No. 108-21
__________
Printed for the use of the Committee on Science
Available via the World Wide Web: http://www.house.gov/science
______
88-231 U.S. GOVERNMENT PRINTING OFFICE
WASHINGTON : 2003
____________________________________________________________________________
For Sale by the Superintendent of Documents, U.S. Government Printing Office
Internet: bookstore.gpo.gov Phone: toll free (866) 512-1800; (202) 512�091800
Fax: (202) 512�092250 Mail: Stop SSOP, Washington, DC 20402�090001
COMMITTEE ON SCIENCE
HON. SHERWOOD L. BOEHLERT, New York, Chairman
LAMAR S. SMITH, Texas RALPH M. HALL, Texas
CURT WELDON, Pennsylvania BART GORDON, Tennessee
DANA ROHRABACHER, California JERRY F. COSTELLO, Illinois
JOE BARTON, Texas EDDIE BERNICE JOHNSON, Texas
KEN CALVERT, California LYNN C. WOOLSEY, California
NICK SMITH, Michigan NICK LAMPSON, Texas
ROSCOE G. BARTLETT, Maryland JOHN B. LARSON, Connecticut
VERNON J. EHLERS, Michigan MARK UDALL, Colorado
GIL GUTKNECHT, Minnesota DAVID WU, Oregon
GEORGE R. NETHERCUTT, JR., MICHAEL M. HONDA, California
Washington CHRIS BELL, Texas
FRANK D. LUCAS, Oklahoma BRAD MILLER, North Carolina
JUDY BIGGERT, Illinois LINCOLN DAVIS, Tennessee
WAYNE T. GILCHREST, Maryland SHEILA JACKSON LEE, Texas
W. TODD AKIN, Missouri ZOE LOFGREN, California
TIMOTHY V. JOHNSON, Illinois BRAD SHERMAN, California
MELISSA A. HART, Pennsylvania BRIAN BAIRD, Washington
JOHN SULLIVAN, Oklahoma DENNIS MOORE, Kansas
J. RANDY FORBES, Virginia ANTHONY D. WEINER, New York
PHIL GINGREY, Georgia JIM MATHESON, Utah
ROB BISHOP, Utah DENNIS A. CARDOZA, California
MICHAEL C. BURGESS, Texas VACANCY
JO BONNER, Alabama
TOM FEENEY, Florida
RANDY NEUGEBAUER, Texas
C O N T E N T S
July 16, 2003
Page
Witness List..................................................... 2
Hearing Charter.................................................. 3
Opening Statements
Statement by Representative Sherwood L. Boehlert, Chairman,
Committee on Science, U.S. House of Representatives............ 14
Written Statement............................................ 15
Statement by Representative Ralph M. Hall, Minority Ranking
Member, Committee on Science, U.S. House of Representatives.... 15
Written Statement............................................ 16
Prepared Statement by Representative Eddie Bernice Johnson,
Member, Committee on Science, U.S. House of Representatives.... 17
Witnesses:
Dr. Raymond L. Orbach, Director, Office of Science, Department of
Energy
Oral Statement............................................... 19
Written Statement............................................ 20
Biography.................................................... 27
Dr. Peter A. Freeman, Assistant Director, Computer and
Information Science and Engineering Directorate, National
Science Foundation
Oral Statement............................................... 28
Written Statement............................................ 30
Biography.................................................... 33
Dr. Daniel A. Reed, Director, National Center for Supercomputing
Applications, University of Illinois at Urbana-Champaign
Oral Statement............................................... 35
Written Statement............................................ 37
Biography.................................................... 41
Financial Disclosure......................................... 42
Mr. Vincent F. Scarafino, Manager, Numerically Intensive
Computing, Ford Motor Company
Oral Statement............................................... 44
Written Statement............................................ 45
Biography.................................................... 47
Financial Disclosure......................................... 48
Discussion....................................................... 49
Appendix 1: Answers to Post-Hearing Questions
Dr. Raymond L. Orbach, Director, Office of Science, Department of
Energy......................................................... 76
Dr. Peter A. Freeman, Assistant Director, Computer and
Information Science and Engineering Directorate, National
Science Foundation............................................. 79
Dr. Daniel A. Reed, Director, National Center for Supercomputing
Applications, University of Illinois at Urbana-Champaign....... 87
Appendix 2: Additional Material for the Record
Dr. Raymond L. Orbach, insert concerning Japan's role in
nanotechnology as it relates to supercomputing................. 92
SUPERCOMPUTING: IS THE U.S. ON THE RIGHT PATH?
----------
WEDNESDAY, JULY 16, 2003
House of Representatives,
Committee on Science,
Washington, DC.
The Committee met, pursuant to call, at 10:21 a.m., in Room
2318 of the Rayburn House Office Building, Hon. Sherwood L.
Boehlert (Chairman of the Committee) presiding.
HEARING CHARTER
COMMITTEE ON SCIENCE
U.S. HOUSE OF REPRESENTATIVES
Supercomputing: Is the U.S.
on the Right Path?
WEDNESDAY, JULY 16, 2003
10:00 A.M.-12:00 P.M.
2318 RAYBURN HOUSE OFFICE BUILDING
1. Purpose
On Wednesday, July 16, 2003, the House Science Committee will hold
a hearing to examine whether the United States is losing ground to
foreign competitors in the production and use of supercomputers\1\ and
whether federal agencies' proposed paths for advancing our
supercomputing capabilities are adequate to maintain or regain the U.S.
lead.
---------------------------------------------------------------------------
\1\ Supercomputing is also referred to as high-performance
computing, high-end computing, and sometimes advanced scientific
computing.
---------------------------------------------------------------------------
2. Witnesses
Dr. Raymond L. Orbach is the Director of the Office of Science at the
Department of Energy. Prior to joining the Department, Dr. Orbach was
Chancellor of the University of California at Riverside.
Dr. Peter A. Freeman is Assistant Director for the Computer and
Information Science and Engineering Directorate (CISE) at the National
Science Foundation (NSF). Prior to joining NSF in 2002, he was
professor and founding Dean of the College of Computing at Georgia
Institute of Technology.
Dr. Daniel A. Reed is the Director of the National Center for
Supercomputing Applications (NCSA) at the University of Illinois at
Urbana-Champaign. NCSA is the leader of one of NSF's two university-
based centers for high-performance computing. Dr. Reed is also the
Director of the National Computational Science Alliance and is a
principal investigator in the National Science Foundation's TeraGrid
project. Earlier this year, Dr. Reed was appointed to the President's
Information Technology Advisory Committee (PITAC).
Mr. Vincent Scarafino is the Manager of Numerically Intensive Computing
at Ford Motor Company, where he focuses on providing flexible and
reliable supercomputer resources for Ford's vehicle product
development, including vehicle design and safety analysis.
3. Overarching Questions
The hearing will address the following overarching questions:
1. Is the U.S. losing its leadership position in
supercomputing? Do the available supercomputers allow United
States science and industry to be competitive internationally?
Are federal efforts appropriately targeted to deal with this
challenge?
2. Are federal agencies pursuing conflicting supercomputing
programs? What can be done to ensure that federal agencies
pursue a coordinated policy for providing supercomputing to
meet the future needs for science, industry, and national
defense?
3. Is the National Science Foundation moving away from the
policies and programs that in the past have provided broad
national access to advanced supercomputers?
4. Can the U.S. fulfill its scientific and defense
supercomputing needs if it continues to rely on machines
designed for mass-market commercial applications?
4. Brief Overview
High-performance computers (also called
supercomputers) are an essential component of U.S. scientific,
industrial, and military competitiveness. However, the fastest
and most efficient supercomputer in the world today is in
Japan, not the U.S. Some experts claim that Japan was able to
produce a computer so far ahead of the American machines
because the U.S. had taken an overly cautious or conventional
approach for developing new high-performance computing
capabilities.
Users of high-performance computing are spread
throughout government, industry, and academia, and different
high-performance computing applications are better suited to
different types of machines. As the U.S. works to develop new
high-performance computing capabilities, extraordinary
coordination among agencies and between government and industry
will be required to ensure that creative new capabilities are
developed efficiently and that all of the scientific,
governmental, and industrial users have access to the high-
performance computing hardware and software best suited to
their applications.
The National Science Foundation (NSF) currently
provides support for three supercomputing centers: the San
Diego Supercomputer Center, the National Center for
Supercomputing Applications at Urbana-Champaign in Illinois,
and the Pittsburgh Supercomputing Center. These centers, along
with their partners at other universities, are the primary
source of high-performance computing for researchers in many
fields of science. Currently, support for these centers beyond
fiscal year 2004 is uncertain, and in the past few years NSF
has been increasing its investment in a nationwide computing
grid, in which fast connections are built between many
computers to allow for certain types of high-performance
scientific computing and advanced communications and data
management. It is not clear whether this ``grid computing''
approach will provide the high-performance computing
capabilities needed in all the scientific fields that currently
rely on the NSF supercomputing centers.
At the Department of Energy, there are two programs
aimed at advancing high-performance computing capabilities.
One, in the National Nuclear Security Administration (NNSA), is
the continuation of a long-term effort to provide
supercomputers to be used for modeling nuclear weapons effects;
these simulations are particularly important in light of
existing bans on nuclear weapon testing. In the other program,
the Office of Science is now proposing to supplement its
current advanced scientific computing activities with a new
effort designed to create the world's fastest supercomputers.
5. Current Issues
Is the U.S. Competitive?
Japan's Earth Simulator is designed to perform simulations of the
global environment that allow researchers to study scientific questions
related to climate, weather, and earthquakes. It was built by NEC for
the Japanese government at a cost of at least $350 million and has been
the fastest computer in the world since it began running in March 2002.
When the first measures of its speed were performed in April 2002,
researchers determined that the Earth Simulator was almost five times
faster than the former record holder, the ASCI White System at Lawrence
Livermore National Laboratory, and also used the machine's computing
power significantly more efficiently.\2\
---------------------------------------------------------------------------
\2\ For the U.S. supercomputers, typical scientific applications
usually only are able to utilize 5-10 percent of the theoretical
maximum computing power, while the design of the Earth Simulator makes
30-50 percent of its power accessible to the majority of typical
scientific applications.
---------------------------------------------------------------------------
This new development caused a great deal of soul-searching in the
high-performance computing community about the U.S. approach to
developing new capabilities and the emphasis on using commercially
available (not specialized or custom-made) components. Going forward,
it is not clear whether or not such a commodity-based approach will
allow the U.S. high-performance computing industry to remain
competitive. It is also unclear if the new machines produced by this
approach will be able provide American academic, industrial, and
governmental users with the high-performance computing capabilities
they need to remain the best in the world in all critical applications.
Will All Users Be Served?
Users of high-performance computing are spread throughout
government, industry, and academia. Different high-performance
computing applications are better suited to different types of
machines. For example, weather modeling and simulations of nuclear
weapons require many closely-related calculations, so machines for
these applications must have components that communicate with each
other quickly and often. Other applications, such as simulations of how
proteins fold, can be efficiently performed with a more distributed
approach on machines in which each component tackles a small piece of
the problem and works in relative isolation. In the U.S., the major
producers of high-performance computers include IBM, Hewlett-Packard,
and Silicon Graphics, Inc., whose products lean toward the more
distributed approach, and Cray, whose products are more suited to
problems that require the performance of closely-related calculations.
The Japanese (NEC, Fujitsu, and Hitachi), also produce this sort of
machine. The concern is that the U.S. on the whole has moved away from
developing and manufacturing the machines needed for problems with
closely-related calculations because the more distributed machines have
a bigger commercial market. The Japanese have been filling this gap,
but the gap could still impact the access of American scientists to the
types of supercomputers that they need for certain important research
problems.
Responsibility for providing high-performance computing
capabilities to existing users and for developing new capabilities is
distributed among 11 different federal agencies and offices and relies
heavily on industry for development and production. In this
environment, extraordinary amounts of coordination are needed to ensure
that new capabilities are developed efficiently and that the most
appropriate kinds of hardware and software are available to the
relevant users--coordination among agencies and between government and
industry, as well as cooperation among universities and hardware and
software companies. The results of an ongoing interagency effort to
produce a coherent high-performance computing roadmap and the influence
this roadmap has on agencies' programs will be the first test.
Where are the DOE Office of Science and the NSF Programs Headed?
Both NSF and the DOE Office of Science are moving ahead in
significant new directions. At NSF, no plans have been announced to
continue the Partnerships for Advanced Computational Infrastructure
program, which supports the supercomputer centers, beyond fiscal year
2004. In addition, a proposed reorganization of NSF's Computer and
Information Sciences and Engineering Directorate was announced on July
9 that includes a merging of the Advanced Computational Infrastructure
program (which includes the support for the supercomputer centers) and
the Advanced Networking Infrastructure program (which supports efforts
on grid computing--an alternative approach to high-performance
computing). Some scientists have expressed concerns that NSF may be
reducing its commitment to providing researchers with a broad range of
supercomputing capabilities and instead focusing its attention on grid
computing and other distributed approaches.
For the DOE Office of Science, the fiscal year 2004 budget request
proposes a new effort in next-generation computer architecture to
identify and address major bottlenecks in the performance of existing
and planned DOE science applications. In addition, the July 8 mark-up
of the House Energy and Water Development Appropriations Subcommittee
sets funding for the Advanced Scientific Computing Research initiative
at $213.5 million, an increase of $40 million over the request and $46
million over the previous year. Decisions about the future directions
for high-performance computing at NSF and DOE Office of Science are
clearly being made now.
The White House has an interagency effort underway, the High End
Computing Revitalization Task Force (HECRTF), which is supposed to
result in the agencies' submitting coordinated budget requests in this
area for fiscal year 2005.
6. Background
What is High-Performance Computing? High-performance computing--also
called supercomputing, high-end computing, and sometimes advanced
scientific computing--is a phrase used to describe machines or groups
of machines that can perform very complex computations very quickly.
These machines are used to solve complicated and challenging scientific
and engineering problems or manage large amounts of data. There is no
set definition of how fast a computer must be to be ``high-
performance'' or ``super,'' as the relevant technologies improve so
quickly that the high-performance computing achievements of a few years
ago could be handled now by today's desktops. Currently, the fastest
supercomputers are able to perform trillions of calculations per
second.
What is High-Performance Computing Used For? High-performance computing
is needed for a variety of scientific, industrial, and national defense
applications. Most often, these machines are used to simulate a
physical system that is difficult to study experimentally. The goal can
be to use the simulation as an alternative to actual experiments (e.g.,
for nuclear weapon testing and climate modeling), as a way to test our
understanding of a system (e.g., for particle physics and
astrophysics), or as a way to increase the efficiency of future
experiments or product design processes (e.g., for development of new
industrial materials or fusion reactors). Other major uses for
supercomputers include performing massive complex mathematical
calculations (e.g., for cryptanalysis) or managing massive amounts of
data (e.g., for government personnel databases).
Scientific Applications: There are a rich variety of
scientific problems being tackled using high-performance
computing. Large-scale climate modeling is used to examine
possible causes and future scenarios related to global warming.
In biology and biomedical sciences, researchers perform
simulations of protein structure, folding, and interaction
dynamics and also model blood flows. Astrophysics model planet
formation and supernova, and cosmologists analyze data on light
from the early universe. Particle physicists use the ultra-fast
computers to perform the complex calculations needed to study
quantum chromodynamics and improve our understanding of
electrons and quarks, the basic building blocks of all matter.
Geologists model the stresses within the earth to study plate
tectonics, while civil engineers simulate the impact of
earthquakes.
National Defense Applications: There are a number of ways in
which high-performance computing is used for national defense
applications. The National Security Agency (NSA) is a major
user and developer of high-performance computers for executing
specialized tasks relevant to cryptanalysis (such as factoring
large numbers). The Department of Energy's National Nuclear
Security Administration is also a major user and developer of
machines to be used for designing and modeling nuclear weapons.
Other applications within the Department of Defense include
armor penetration modeling, weather forecasting, and
aerodynamics modeling. Many of the scientific applications also
have direct or future defense applications. For example,
computational fluid dynamics studies are also of interest to
the military, e.g. for modeling turbulence around aircraft. The
importance of high-performance computing in many military
areas, including nuclear and conventional weapons design, means
that machines that alone or when wired together are capable of
superior performance at military tasks are subject to U.S.
export controls.
Industrial Applications: Companies use high-performance
computing in a variety of ways. The automotive industry uses
fast machines to maximize the effectiveness of computer-aided
design and engineering. Pixar uses massive computer animation
programs to produce films. Pharmaceutical companies simulate
chemical interactions to help with drug design. The commercial
satellite industry needs to manage huge amounts of data for
mapping. Financial companies and other industries use large
computers to process immense and unpredictable Web transaction
volumes, mine databases for sales patterns or fraud, and
measure the risk of complex investment portfolios.
What Types of High-Performance Computers Are There? All of the above
examples of high-performance computing applications require very fast
machines, but they do not all require the same type of very fast
machine. There are a number of different ways to build high-performance
computers, and different configurations are better suited to different
problems. There are many possible configurations, but they can be
roughly divided into two classes: big, single-location machines and
distributed collections of many computers (this approach is often
called grid computing). Each approach has its benefits--the big
machines can be designed for a specific problem and are often faster,
while grid computing is attractive in part because by using a multitude
of commercially-available computers, the purchase and storage cost is
often lower than for a large specialized supercomputer.
Since the late 1990's, the U.S. approach to developing new
capabilities has emphasized using commercially available (not
specialized) components as much as possible. This emphasis has resulted
in an increased focus on grid computing, and, in large machines, has
led to a hybrid approach in which companies use commercial processors
(whose speed is increasing rapidly anyway) to build the machines and
then further speed them up by increasing the number of processors and
improving the speed at which information is passed between processors.
There are a number of distinctions that can be made among large
machines bases on how the processors are connected. The differences
relate to how fast and how often the various components of the computer
communicate with each other and how calculations are distributed among
the components.
Users thus have a number of options for high-performance computing.
Each user must take into account all of the pros and cons of the
different configurations when he is deciding what sort of machine to
use and how to design software to allow that machine to most
efficiently solve his problem. For example, some problems, like weather
and climate modeling and cryptanalysis, require lots of communication
among computer components and large quantities of stored data, while
other applications, like large-scale data analysis for high energy
physics experiments or bioinformatics projects, can be more efficiently
performed on distributed machines each tackling its own piece of the
problem in relative isolation.
How Do Government and Industry Provide Existing and New High-
Performance Computing Capabilities? The development and production of
high-performance computing capabilities requires significant effort by
both government and industry. For any of the applications of high-
performance computing described above, the users need good hardware
(the high-performance machine or group of machines) and good software
(programs that allow them to perform their calculations as accurately
and efficiently as possible).
The role of government therefore includes (1) funding research on
new approaches to building high-performance computing hardware, (2) in
some cases, funding the development stage of that hardware (usually
through security agencies), (3) purchasing the hardware to be used by
researchers at universities and personnel at government agencies, (4)
funding research on software and programs to use existing and new high-
performance computing capabilities, and (5) supporting research that
actually uses the hardware and software. The role of industry is
complementary--i.e., it receives funding to do research and development
on new hardware and software, and it is the seller of this hardware and
software to government agencies, universities, and companies. The
primary industries involved in producing high-performance computing
capabilities are computer makers (such as IBM, Hewlett-Packard, Silicon
Graphics, Inc., and Cray), chip makers (such as Intel), and software
designers. Congress has long had concerns about the health of the U.S.
supercomputing industry. In 1996, when the National Center for
Atmospheric Research, a privately-run, federally-funded research
center, tried to order a supercomputer from NEC for climate modeling,
Congress blocked the purchase.
Federal High-Performance Computing Programs: In 1991, Congress passed
the High Performance Computing Act, establishing an interagency
initiative (now called National Information Technology Research and
Development (NITRD) programs) and a National Coordination Office for
this effort. Currently 11 agencies or offices participate in the high-
end computing elements of the NITRD program (See Table 1 in the
appendix). The total requested by all 11 agencies in fiscal year 2003
for high-end computing was $846.5 million. The largest research and
development programs are at the National Science Foundation (NSF),
which requested $283.5 million, and the Department of Energy Office of
Science, which requested $137.8 million. Other major agency activities
(all between $80 and $100 million) are at the National Institutes of
Health (NIH), the Defense Advanced Research Projects Agency (DARPA),
the National Aeronautics and Space Administration (NASA), and the
Department of Energy's National Nuclear Security Administration (NNSA).
Different agencies concentrate on serving different user communities
and on different stages of hardware and software development and
application. (In addition to the research and development-type
activities that are counted for the data included in Table 1 and
referenced above, many agencies, such as NNSA and the National Oceanic
and Atmospheric Administration (NOAA), devote significant funding to
the purchase and operation of high-performance computers that perform
these agencies' mission-critical applications.) \3\
---------------------------------------------------------------------------
\3\ For example, in FY 2003 NOAA spent $36 million on
supercomputers--$10 million for machines for climate modeling and $26
million for machines for the National Weather Service.
National Science Foundation: The NSF serves a very wide
variety of scientific fields within the academic research
community, mainly through a series of supercomputing centers,
originally established in 1985 and currently funded under the
Partnerships for Advanced Computational Infrastructure (PACI)
program. The supercomputer centers provide researchers not only
with access to high-performance computing capabilities but also
with tools and expertise on how best to utilize these
resources. The NSF also is supporting the development of the
Extensible Terascale Facility (ETF), a nationwide grid of
machines that can be used for high-performance computing and
advanced communications and data management. Recently, some
researchers within the high-performance computing community
have expressed concern that NSF may be reducing its commitment
to the supercomputer centers and increasing its focus on grid
computing and distributed approaches to high-performance
---------------------------------------------------------------------------
computing, such as would be used in the ETF.
Department of Energy: The Department of Energy has been a
major force in advancing high-performance computing for many
years, and the unveiling of the fastest computer in the world
in Japan in 2002 resulted in serious self-evaluation at the
department, followed by a rededication to efforts to enhance
U.S. supercomputing capabilities. The Department of Energy has
two separate programs focused on both developing and applying
high-performance computing. The Advanced Scientific Computing
Research (ASCR) program in the Office of Science funds research
in applied mathematics (to develop methods to model complex
physical and biological systems), in network and computer
sciences, and in advanced computing software tools. For fiscal
year 2004, the department has proposed a new program on next-
generation architectures for high-performance computing. The
Accelerated Strategic Computing Initiative (ASCI) is part of
the NNSA's efforts to provide advanced simulation and computing
technologies for weapons modeling.
DARPA: DARPA traditionally focuses on the development of new
hardware, including research into new architectures and early
development of new systems. On July 8, DARPA announced that
Cray, IBM, and Sun Microsystems had been selected as the three
contractor teams for the second phase of the High Productivity
Computing Systems program, in which the goal is to provide a
new generation of economically viable, scalable, high
productivity computing systems for the national security and
industrial user communities in the 2009 to 2010 timeframe.
Other Agencies: NIH, NASA, and NOAA are all primarily users of
high performance computing. NIH manages and analyzes biomedical
data and models biological processes. NOAA uses simulations to
do weather forecasting and climate change modeling. NASA has a
variety of applications, including atmospheric modeling,
aerodynamic simulations, and data analysis and visualization.
The National Security Agency (NSA) both develops and uses high-
performance computing for a number of applications, including
cryptanalysis. As a user, NSA has a significant impact on the
high-performance computing market, but due to the classified
nature of its work, the size of its contributions to High End
Computing Infrastructure and Applications and the amount of
funding it uses for actual operation of computers is not
included in any of the data.
Interagency Coordination: The National Coordination Office (NCO)
coordinates planning, budget, and assessment activities for the Federal
Networking and NITRD Program through a number of interagency working
groups. The NCO reports to the White House Office of Science and
Technology Policy and the National Science and Technology Council. In
2003, NCO is also managing the High End Computing Revitalization Task
Force (HECRTF), an interagency effort on the future of U.S. high-
performance computing. The HECRTF is tasked with development of a
roadmap for the interagency research and development for high-end
computing core technologies, a federal high-end computing capacity and
accessibility improvement plan, and a discussion of issues relating to
federal procurement of high-end computing systems. The product of the
HECRTF process is expected to guide future investments in this area,
starting with agency budget submissions for fiscal year 2005.
The Role of Industry: Industry plays a critical role in developing and
providing high-performance computing capabilities to scientific,
industrial, and defense users. Many supercomputers are purchased
directly from computer companies like IBM, Hewlett-Packard, Silicon
Graphics, Inc., and Cray, and the groups that do build their own high-
performance clusters do so from commercially available computers and
workstations. Industry is a recipient of federal funding for initial
research into new architectures for hardware, for development of new
machines, and for production of standard and customized systems for
government and universities, but industry also devotes its own funding
to support research and development. The research programs do not just
benefit the high-performance computing community, as new architectures
and faster chips lay the groundwork for better performing computers and
processors in all commercial information technology products.
The State of the Art in High-Performance Computing: Twice a year, a
list of the 500 fastest supercomputers is compiled; the latest list was
released on June 23, 2003 (see Table 2 in the appendix).\4\ The Earth
Simulator supercomputer, built by NEC and installed last year at the
Earth Simulator Center in Yokohama, Japan, continues to hold the top
spot as the best performer. It is approximately twice as fast as the
second place machine, the ASCI Q system at Los Alamos National
Laboratory, built by Hewlett-Packard. Of the top twenty machines, eight
are located at various Department of Energy national laboratories and
two at U.S. universities,\5\ and nine were made by IBM and five by
Hewlett-Packard.
---------------------------------------------------------------------------
\4\ The top 500 list is compiled by researchers at the University
of Mannheim (Germany), Lawrence Berkeley National Laboratory, and the
University of Tennessee and is available on line at http://
www.top500.org/. For a machine to be included on this public list, its
owners must send information about its configuration and performance to
the list-keepers. Therefore, the list is not an entirely comprehensive
picture of the high-performance computing world, as classified
machines, such as those used by NSA, are not included.
\5\ The two university machines are located at the Pittsburgh
Supercomputing Center (supported primarily by NSF) and Louisiana State
University's Center for Applied Information Technology and Learning.
The remaining 12 machines include four in Europe, two in Japan, and one
each at the National Oceanic & Atmospheric Administration, the National
Center for Atmospheric Research, the Naval Oceanographic Office, and
NASA.
---------------------------------------------------------------------------
7. Witness Questions
The witnesses were asked to address the following questions in
their testimony:
Questions for Dr. Raymond L. Orbach
The Office of Science appears to have embarked on a
new effort in next-generation advanced scientific computer
architecture that differs from the development path currently
pursued by the National Nuclear Security Agency (NNSA), the
lead developer for advanced computational capability at the
Department of Energy (DOE). Why is the Office of Science taking
this approach?
How is the Office of Science cooperating with the
Defense Advanced Research Projects Agency, which supports the
development of advanced computers for use by the National
Security Agency and other agencies within the Department of
Defense?
To what extent will the Office of Science be guided
by the recommendations of the High-End Computing Revitalization
Task Force? How will the Office of Science contribute to the
Office of Science and Technology Policy plan to revitalize
high-end computing?
To what extent are the advanced computational needs
of the scientific community and of the private sector
diverging? What is the impact of any divergence on the advanced
computing development programs at the Office of Science?
Questions for Dr. Peter A. Freeman
Some researchers within the computer science
community have suggested that the NSF may be reducing its
commitment to the supercomputer centers. Is this the case? To
what extent does the focus on grid computing represent a move
away from providing researchers with access to the most
advanced computing equipment?
What are the National Science Foundation's (NSF's)
plans for funding the supercomputer centers beyond fiscal year
2004? To what extent will you be guided by the recommendation
of the NSF Advisory Panel on Cyberinfrastructure to maintain
the Partnerships for Advanced Computational Infrastructure,
which currently support the supercomputer centers?
To what extent will NSF be guided by the
recommendations of the High-End Computing Revitalization Task
Force? How will NSF contribute to the Office of Science and
Technology Policy plan to revitalize high-end computing?
To what extent are the advanced computational needs
of the scientific community and of the private sector
diverging? What is the impact of any such divergence on the
advanced computing programs at NSF?
Questions for Dr. Daniel A. Reed
Some researchers within the computer science
community have suggested that the National Science Foundation
(NSF) may be reducing its commitment to provide advanced
scientific computational capability to U.S. scientists and
engineers. Have you detected any change in policy on the part
of NSF?
What advanced computing capabilities must the Federal
Government provide the academic research community for the
government's programs to be considered successful? Are the
programs for developing the next-generation of advanced
scientific computing that are currently underway at government
agencies on track to provide these capabilities? If not, why
not?
For academic scientists and engineers, what is the
difference between the advanced scientific computing
capabilities provided by NSF and those provided by the
Department of Energy?
Questions for Mr. Vincent F. Scarafino
How does Ford use high-performance computing? How do
computing capabilities affect Ford's competitiveness nationally
and internationally?
What does Ford see as the role of the Federal
Government in advancing high-performance computing capabilities
and in making these capabilities accessible to users? Are
current agency programs for developing the next-generation of
advanced scientific computing adequate to provide these
capabilities? If not, why not?
Is the U.S. government cooperating appropriately with
the private sector on high-performance computing, and is the
level of cooperation adequate to sustain leadership and meet
scientific and industrial needs?
To what extent are the advanced computational needs
of the scientific community and of the private sector
diverging? What is the impact of any divergence on Ford's
access to advanced computing capabilities?
Chairman Boehlert. The hearing will come to order. And I
apologize for being the delinquent Member of the group gathered
here. My apologies to all.
It is a pleasure to welcome everyone here this morning for
this important hearing. At first blush, today's topic,
supercomputing, may seem technical and arcane, of interest to
just a few researchers who spend their lives in the most
rarefied fields of science. But in reality, the subject of this
hearing is simple and accessible, and it has an impact on all
of us, because supercomputing affects the American economy and
our daily lives, perhaps more so than so many, many other
things that we focus a lot of time and attention on.
Supercomputers help design our cars, predict our weather,
and deepen our understanding of the natural forces that govern
our lives, such as our climate. Indeed, computation is now
widely viewed as a third way of doing science; building on the
traditional areas of theory and experimentation.
So when we hear that the U.S. may be losing its lead in
supercomputing, that Japan now has the fastest supercomputer,
that the U.S. may be returning to a time when our top
scientists didn't have access to the best machines, that our
government may have too fragmented a supercomputing policy,
well, those are issues a red flag should be waved on to capture
the attention of all of us.
And those issues have captured our attention. The purpose
of this hearing is to gauge the state of U.S. supercomputing
and to determine how to deal with any emerging problems.
I don't want to exaggerate, we are not at a point of
crisis. Most of the world's supercomputers are still made by
and used by Americans, but we are at a pivotal point when we
need to make critical decisions to make sure that remains the
case.
And maintaining U.S. leadership requires a coordinated,
concerted effort by the Federal Government. Let me stress that.
Coordinated, concerted effort by the Federal Government. The
Federal Government has long underwritten the basic research
that fuels the computer industry, has purchased the highest end
computers, and has ensured that those computers are available
to a wide range of American researchers. This Committee has
played an especially crucial role in ensuring access, pushing
for the creation of the National Science Foundation
Supercomputer Centers back in the early '80's.
Government action is just as needed now. But what action?
The Department of Energy is proposing to move away from our
reliance on more mass-market supercomputers to pursue research
on massive machines designed to solve especially complex
problems. NSF appears to be moving away from super--supporting
supercomputer centers to a more distributed computing approach.
These policies need to be examined.
So with that in mind, here are some of the questions we
intend to pursue today. Is the U.S. losing its lead in
supercomputing, and what can be done about it? What federal
policies should be pursued to maintain our lead, and how should
we judge whether they are succeeding? Is federal policy
sufficiently coordinated? And I think the answer is clear. And
are the new directions being pursued by NSF and the Department
of Energy the proper approach to maintaining our lead?
We have a distinguished group of experts, and I look
forward to hearing their testimony.
With that, it is my pleasure to yield to the distinguished
Ranking Member, Mr. Hall of Texas.
[The prepared statement of Mr. Boehlert follows:]
Prepared Statement of Chairman Sherwood L. Boehlert
It's a pleasure to welcome everyone here this morning for this
important hearing. At first blush, today's topic, supercomputing, may
seem technical and arcane--of interest to just a few researchers who
spend their lives in the most rarefied fields of science. But in
reality, the subject of this hearing is simple and accessible, and it
has an impact on all of us because supercomputing affects the American
economy and our daily lives.
Supercomputers help design our cars, predict our weather, and
deepen our understanding of the natural forces that govern our lives,
such as our climate. Indeed, computation is now widely viewed as a
third way of doing science--building on the traditional areas of theory
and experimentation.
So when we hear that the U.S. may be losing its lead in
supercomputing, that Japan now has the fastest supercomputer, that the
U.S. may be returning to a time when our top scientists didn't have
access to the best machines, that our government may have too
fragmented a supercomputing policy--well, those issues are a red flag
that should capture the attention of all of us.
And those issues have captured our attention. The purpose of this
hearing is to gauge the state of U.S. supercomputing and to determine
how to deal with any emerging problems.
I don't want to exaggerate--we're not at a point of crisis--most of
the world's supercomputers are still made by, and used by Americans.
But we are at a pivotal point when we need to make critical decisions
to make sure that remains the case.
And maintaining U.S. leadership requires a coordinated, concerted
effort by the Federal Government. The Federal Government has long
underwritten the basic research that fuels the computer industry, has
purchased the highest-end computers, and has ensured that those
computers are available to a wide range of American researchers. This
committee has played an especially crucial role in ensuring access,
pushing for the creation of the National Science Foundation (NSF)
Supercomputer Centers back in the early `80s.
Government action is just as needed now. But what action? The
Department of Energy is proposing to move away from our reliance on
more mass-market supercomputers to pursue research on massive machines
design to solve especially complex problems. NSF appears to be moving
away from supporting supercomputer centers to a more distributed
computing approach. These policies need to be examined.
So, with that in mind, here are some of the questions we intend to
pursue today:
Is the U.S. losing its lead in supercomputing and
what can be done about that?
What federal policies should be pursued to maintain
our lead and how should we judge whether they are succeeding?
Is federal policy sufficiently coordinated and are
the new directions being pursued by NSF and the Department of
Energy the proper approach to maintaining our lead?
We have a distinguished group of experts, and I look forward to
hearing their testimony.
Mr. Hall. Mr. Chairman, thank you. And I am pleased to join
you today in welcoming our witnesses. And I thank you for your
time, not only in your appearing here, but in preparation and
travel. And thank you for your usual courtesy.
Computation has become one of the, I guess, principal
tools, along with the theory and experiment for conducting
science and engineering research and development. There is no
question that the U.S. preeminence in science and technology
will not and cannot continue unless our scientists and
engineers have access to the most powerful computers available.
The Science Committee has had a deep and sustained interest
in this subject since the emergence of supercomputing in the
late 1970's. And the initial concern of the Committee was to
ensure that the U.S. scientists and engineers, outside of the
classified research world, had access to the most powerful
computers. We have supported programs to provide this access,
such as the supercomputer centers program at NSF.
Moreover, the Committee has encouraged the efforts of the
Federal R&D agencies to develop a coordinated R&D program to
accelerate computing and networking developments. The High
Performance Computing Act of 1991 formalized this interagency
R&D planning and coordination process.
The value and importance of the resulting interagency
information technology R&D program is quite evident from its
appearances as a formal presidential budget initiative through
three different presidential administrations.
So today, I think we want to assess a particular component
of the federal information technology R&D effort. That is, the
pathway being followed for the development of high-end
computers and the provision being made to provide access to
these machines by the U.S. research community.
Questions have been raised as to whether we put all of our
eggs in one basket by mainly focusing on supercomputers based
on commodity components. The recent success of the specialized
Japanese Earth Simulator computer has triggered a review of the
computing needs of the scientific and technical community and
reconsideration of the R&D and acquisition plan needed for the
next several years and for the longer-term.
So we would be very interested today in hearing from our
witnesses about where we are now in terms of high-end computing
capabilities and where we should be going to provide the kinds
of computer systems needed to tackle the most important and
certainly challenging of all problems.
I also want to explore what the roles of the various
federal agencies ought to be in the development of new classes
of high-end computers and for providing access for the general
U.S. research community to these essential tools.
I appreciate the attendance of our witnesses, and I look
forward to your discussion.
I yield back my time.
[The prepared statement of Mr. Hall follows:]
Prepared Statement of Representative Ralph M. Hall
Mr. Chairman, I am pleased to join you today in welcoming our
witnesses, and I congratulate you on calling this hearing on federal
R&D in support of high-performance computing.
Computation has become one of the principal tools, along with
theory and experiment, for conducting science and engineering research
and development. There is no question that U.S. preeminence in science
and technology will not continue unless our scientists and engineers
have access to the most powerful computers available.
The Science Committee has had a deep and sustained interest in this
subject since the emergence of supercomputing in the late 1970s.
The initial concern of the Committee was to ensure that U.S.
scientists and engineers, outside of the classified research world, had
access to the most powerful computers. we have supported programs to
provide this access, such as the supercomputer centers program at NSF.
Moreover, the Committee has encouraged the efforts of the federal
R&D agencies to develop a coordinated R&D program to accelerate
computing and networking developments. The High Performance Computing
Act of 1991 formalized this interagency R&D planning and coordination
process.
The value and importance of the resulting interagency information
technology R&D program is evident from its appearance as a formal
presidential budget initiative through 3 different Administrations.
Today, we want to assess a particular component of the federal
information technology R&D effort. That is, the pathway being followed
for the development of high-end computers and the provisions being made
to provide access to these machines by the U.S. research community.
Questions have been raised as to whether we have put all of our
eggs into one basket by mainly focusing on supercomputers based on
commodity components. The recent success of the specialized Japanese
Earth Simulator computer has triggered a review of the computing needs
of the scientific and technical community and a reconsideration of the
R&D and acquisition plan needed for the next several years, and for the
longer-term.
I will be interested in hearing from our witnesses about where we
are now in terms of high-end computing capabilities and where we should
be going to provide the kinds of computer systems needed to tackle the
most important and computationally challenging problems.
I also want to explore what the roles of the various federal
agencies ought to be in the development of new classes of high-end
computers and for providing access for the general U.S. research
community to these essential tools.
I appreciate the attendance of our witnesses, and I look forward to
our discussion.
Chairman Boehlert. Thank you very much, Mr. Hall. And
without objection, all other's opening statements will be made
a part of the record at this juncture.
[The prepared statement of Ms. Johnson follows:]
Prepared Statement of Representative Eddie Bernice Johnson
Thank you, Chairman for calling hearing to examine the very
important issue of Supercomputing. I also want to thank our witnesses
for agreeing to appear today.
We are here to discuss whether the United States is losing ground
to foreign competitors in the production and use of supercomputers and
whether federal agencies' proposed paths for advancing our
supercomputing capabilities are adequate to maintain or regain the U.S.
lead.
As we all know, a supercomputer is a broad term for one of the
fastest computers currently available. Such computers are typically
used for number crunching including scientific simulations, (animated)
graphics, analysis of geological data (e.g., in petrochemical
prospecting), structural analysis, computational fluid dynamics,
physics, chemistry, electronic design, nuclear energy research and
meteorology.
Supercomputers are state-of-the-art, extremely powerful computers
capable of manipulating massive amounts of data in a relatively short
time. They are very expensive and are employed for specialized
scientific and engineering applications that must handle very large
databases or do a great amount of computation, among them meteorology,
animated graphics, fluid dynamic calculations, nuclear energy research
and weapon simulation, and petroleum exploration.
Supercomputers are gaining popularity in all corners of corporate
America. They are used to analyze vehicle crash test by auto
manufacturers, evaluate human diseases and develop treatments by the
pharmaceutical industry and test aircraft engines by the aero-space
engineers.
It quite evident that supercomputing will become more important to
America's commerce in the future. I look forward to working with this
committee on its advancement. Again, I wish thank the witnesses for
coming here today help us conceptualize this goal.
Chairman Boehlert. And just a little history. I can recall
back in the early '80's, 1983 to be exact, when I was a
freshman and Mr. Hall was an emerging power in the Congress. I
sat way down in the front row on the end, and I didn't know
what was going on. But I do remember very vividly the testimony
of a Nobel Laureate, Dr. Ken Wilson, who at that time was at
Cornell University. And he told us that the typical graduate
student in the United Kingdom or Japan or Germany--the typical
graduate student had greater access to the latest in computer
technology than did he, a young Nobel Laureate, you know, one
of the great resources of our nation.
And he argued very forcibly and very persuasively for the
Federal Government to get more actively involved. And boy,
that--it was like the light bulb going on. I didn't even
understand it, and I am not quite sure I do yet, this--the--all
of the intricacies of this supercomputer technology, but I do
remember then being a champion from day one getting this
supercomputer initiative going for America. And in '85, NSF set
up the Supercomputing Centers and--at Carnegie Mellon and
Cornell and others--and boy, did we just go forward, leapfrog
ahead. We did wonderful things.
You know what? A little lethargy is setting in and I am
getting concerned. I am deeply concerned, and I mention in my
opening statement about five times, I should have mentioned it
about 55 times, the importance of a well coordinated federal
response to this issue. I don't want to be second to anybody,
neither does Mr. Hall, neither do any Members of this
committee.
We have an opportunity, and we are going to seize it. And
so we are looking at all of you as resources for this
committee. You are all very distinguished in your own way. You
are very knowledgeable. You will share with us, and hopefully
we will get a few more light bulbs turned on up here. And we
can go forward together. There is an awful lot at stake.
And so I look forward to your testimony, and I hope that
you will sign on here and now. Some of you have no choice. You
have to, right? But sign on here and now to work cooperatively
with us, because there is so much at stake.
With that, let me introduce our witnesses.
Witness consist--list consists of Dr. Raymond Orbach,
Director of the Office of Science, Department of Energy. Dr.
Orbach, good to have you back. Dr. Peter A. Freeman, Assistant
Director, Computer and Information Science and Engineering
Directorate at the National Science Foundation. It is good to
see you once again. Dr. Daniel A. Reed, Director, National
Center for Supercomputing Applications, University of Illinois
at Urbana-Champaign. Dr. Reed. And Mr. Vincent Scarafino,
Manager, Numerically Intensive Computing, Ford Motor Company.
Mr. Scarafino, glad to have you here.
We would ask all of our witnesses to try to summarize your
complete statement, because we all have the benefit of the
complete statement. And we will read those statements very
carefully, but try to summarize in five minutes or so. I am not
going to be arbitrary. This is too darn important to restrict
your expert input to 300 seconds, but I would ask you to be
close to the five minutes. And then we can have a good exchange
in the dialogue. And hopefully we will all profit from this
little exercise we are engaged in.
Dr. Orbach.
STATEMENT OF DR. RAYMOND L. ORBACH, DIRECTOR, OFFICE OF
SCIENCE, DEPARTMENT OF ENERGY
Dr. Orbach. Chairman Boehlert and Ranking Member Hall,
Members of the Committee, I commend you for holding this
hearing. And I deeply appreciate the opportunity to testify on
behalf of the Office of Science at the Department of Energy on
a subject of central importance to this Nation, as you have
both outlined, our need for advanced supercomputing capability.
Through the efforts of the DOE's Office of Science and other
federal agencies, we are working to develop the next generation
of advanced scientific computational capacity, a capability
that supports economic competitiveness and America's scientific
enterprise.
The Bush Administration has forged an integrated and
unified interagency road map to the critical problems that you
have asked us to address today. In my opening statement, I
would like to briefly address the four specific questions that
the Committee has asked of me, and more extensive answers are
contained in my written testimony.
The first question that the Committee addressed to me
concerned the development path for a next-generation advanced
scientific computer and whether the Office of Science path
differed from that of the National Nuclear Security Agency, the
NNSA. The NNSA has stewardship responsibilities and the
computer architectures, which they have used, are well suited
for those needs. And indeed, they led the way in the massively
parallel machine development.
However, those machines operate at only something like five
to 10 percent efficiency when applied to many problems of
scientific interest and also industrial interest. And that
reduces the efficiencies of these high peak speed machines.
Other architectures have shown efficiencies closer to 60
percent for some of the physical problems that science and
industry must address.
We are working with NNSA as partners to explore these
alternatives, which we believe, will be important to both of
our areas of responsibility. For example, the Office of Science
will be exploring computer architectures that may be of value
for magnetic fusion to biology. NNSA is working with us as a
partner to explore equations of state under high pressures and
extreme temperatures, which, of course, is critical to
stewardship issues.
The second question that I was asked was are we cooperating
with DARPA, the Defense Advanced Research Project Agency, and
how does that relationship work. DARPA has historically
invested in new architectures, which have been and are of great
interest to the Office of Science. We are--we have a memorandum
of understanding [MOU] currently under review between the
Defense Department and the Department of Energy that
establishes a framework for cooperation between DARPA and the
DOE, including both the Office of Science and NNSA.
The MOU will cover high-end computation performance
evaluation, development of benchmarks, advanced computer
architecture evaluation, development of mathematical libraries,
and system software. This will bring together the complementary
strengths of each agency. We will be able to draw on DARPA's
strengths in advanced computer architectures and they on our
decades of experience in evaluating new architectures and
transforming them into tools for scientific discovery.
The third question you asked was to what extent will the
Office of Science be guided by the recommendations of the High-
End Computing Revitalization Task Force and how will we
contribute to the OSTP, Office of Science and Technology
Policy, plan to revitalize high-end computation. The formation
of the High-End Computation Revitalization Task Force by OSTP
emphasizes the importance that the Administration places on the
need for a coordinated approach to strengthening high-end
computation. A senior official of the Office of Science is co-
chairing that task force, and it includes many representatives
from across the government and industry.
Many of the task force findings and plans are actually
based on Office of Science practices in advanced computing and
simulation. We are working very closely with NNSA, the
Department of Defense, NASA, the National Science Foundation,
and the National Institutes of Health to assess how best to
coordinate and leverage our agency's high-end computation
investments now and in the future. We expect to play a major
role in executing the plans that emerge from this task force in
partnership with the other agencies and under the guidance of
the President's Science Advisor.
The last question you asked was how are the advanced
computational needs of the scientific community and of the
private sector diverging and how does that affect advanced
computing development programs at the Office of Science. I
don't believe there is a major divergence between the needs of
the scientific community and those of industry's design
engineers. The apparent divergence stems from the dominance of
computer development by the needs of specific commercial
applications: payroll, management information systems, and web
servers.
I will defer to Mr. Scarafino on this, but my own
discussions with industry leaders suggest that the type of
computer architectures that would meet the needs of the Office
of Science would also support their requirements. It would give
them the ability to create virtual prototypes of complex
systems, allowing engineers to optimize different design
parameters without having to build prototypes. This would
reduce the time to market. It would decrease the costs, and it
would increase economic competitiveness.
These four questions were central, and I appreciate being
asked them and given the opportunity to respond. I am gratified
that this committee is intent on enabling us to pursue so many
important computational opportunities for the sake of
scientific discovery, technological innovation, and economic
competitiveness.
I thank you for inviting me, and I will be pleased to take
questions.
[The prepared statement of Dr. Orbach follows:]
Prepared Statement of Raymond L. Orbach
Mr. Chairman and Members of the Committee, I commend you for
holding this hearing--and I appreciate the opportunity to testify on
behalf of the Department of Energy's (DOE) Office of Science--on a
subject of central importance to this nation: our need for advanced
supercomputing capability. Through the efforts of DOE's Office of
Science and other federal agencies, we are working to develop the next
generation of advanced scientific computational capability, a
capability that supports economic competitiveness and America's
scientific enterprise.
As will become abundantly clear in my testimony, the Bush
Administration has forged an integrated and unified interagency roadmap
to the critical problems you have asked us to address today. No one
agency can--or should--carry all the weight of ensuring that our
scientists have the computational tools they need to do their job, yet
duplication of effort must be avoided. The President, and John
Marburger, Office of Science and Technology Policy Director, understand
this. That is why all of us here are working as a team on this problem.
* * *
Mr. Chairman, for more than half a century, every President and
each Congress has recognized the vital role of science in sustaining
this nation's leadership in the world. According to some estimates,
fully half of the growth in the U.S. economy in the last 50 years stems
from federal funding of scientific and technological innovation.
American taxpayers have received great value for their investment in
the basic research sponsored by the Office of Science and other
agencies in our government.
Ever since its inception as part of the Atomic Energy Commission
immediately following World War II, the Office of Science has blended
cutting edge research and innovative problem solving to keep the U.S.
at the forefront of scientific discovery. In fact, since the mid-
1940's, the Office of Science has supported the work of more than 40
Nobel Prize winners, testimony to the high quality and importance of
the work it underwrites.
Office of Science research investments historically have yielded a
wealth of dividends including: significant technological innovations;
medical and health advances; new intellectual capital; enhanced
economic competitiveness; and improved quality of life for the American
people.
Mr. Chairman and Members of this committee, virtually all of the
many discoveries, advances, and accomplishments achieved by the Office
of Science in the last decade have been underpinned by advanced
scientific computing and networking tools developed by the Office of
Advanced Scientific Computing Research (ASCR).
The ASCR program mission is to discover, develop, and deploy the
computational and networking tools that enable scientific researchers
to analyze, model, simulate, and predict complex phenomena important to
the Department of Energy--and to the U.S. and the world.
In fact, by fulfilling this mission over the years, the Office of
Science has played a leading role in maintaining U.S. leadership in
scientific computation worldwide. Consider some of the innovations and
contributions made by DOE's Office of Science:
helped develop the Internet;
pioneered the transition to massively parallel
supercomputing in the civilian sector;
began the computational analysis of global climate
change;
developed many of the DNA sequencing and
computational technologies that have made possible the
unraveling of the human genetic code; and
opened the door for major advances in nanotechnology
and protein crystallography.
* * *
Computational modeling and simulation are among the most
significant developments in the practice of scientific inquiry in the
latter half of the 20th Century. In the past century, scientific
research has been extraordinarily successful in identifying the
fundamental physical laws that govern our material world. At the same
time, the advances promised by these discoveries have not been fully
realized, in part because the real-world systems governed by these
physical laws are extraordinarily complex. Computers help us to
visualize, to test hypotheses, to guide experimental design, and most
importantly to determine if there is consistency between theoretical
models and experiment. Computer-based simulation provides a means for
predicting the behavior of complex systems that can only be described
empirically at present. Since the development of digital computers in
mid-century, scientific computing has greatly advanced our
understanding of the fundamental processes of nature, e.g., fluid flow
and turbulence in physics, molecular structure and reactivity in
chemistry, and drug-receptor interactions in biology. Computational
simulation has even been used to explain, and sometimes predict, the
behavior of such complex natural and engineered systems as weather
patterns and aircraft performance.
Within the past two decades, scientific computing has become a
contributor to essentially all scientific research programs. It is
particularly important to the solution of research problems that are
(i) insoluble by traditional theoretical and experimental approaches,
e.g., prediction of future climates or the fate of underground
contaminants; (ii) hazardous to study in the laboratory, e.g.,
characterization of the chemistry of radionuclides or other toxic
chemicals; or (iii) time-consuming or expensive to solve by traditional
means, e.g., development of new materials, determination of the
structure of proteins, understanding plasma instabilities, or exploring
the limitations of the ``Standard Model'' of particle physics. In many
cases, theoretical and experimental approaches do not provide
sufficient information to understand and predict the behavior of the
systems being studied. Computational modeling and simulation, which
allows a description of the system to be constructed from basic
theoretical principles and the available experimental data, are keys to
solving such problems.
Advanced scientific computing is indispensable to DOE's missions.
It is essential to simulate and predict the behavior of nuclear
weapons, accelerate the development of new energy technologies, and the
aid in discovery of new scientific knowledge.
As the lead government funding agency for basic research in the
physical sciences, the Office of Science has a special responsibility
to ensure that its research programs continue to advance the frontiers
of science. All of the research programs in DOE's Office of Science--in
Basic Energy Sciences, Biological and Environmental Research, Fusion
Energy Sciences, and High-Energy and Nuclear Physics--have identified
major scientific questions that can only be addressed through advances
in scientific computing. This will require significant enhancements to
the Office of Science's scientific computing programs. These include
both more capable computing platforms and the development of the
sophisticated mathematical and software tools required for large scale
simulations.
Existing highly parallel computer architectures, while extremely
effective for many applications, including solution of some important
scientific problems, are only able to operate at 5-10 percent of their
theoretical maximum capability on other applications. Therefore, we
have initiated a Next Generation Architecture program to evaluate the
effectiveness of various different computer architectures in
cooperation with the National Nuclear Security Administration (NNSA)
and the Defense Advanced Research Project Agency to identify those
architectures which are most effective in addressing specific types of
simulations.
To address the need for mathematical and software tools, and to
develop highly efficient simulation codes for scientific discovery, the
Office of Science launched the Scientific Discovery through Advanced
Computing (SciDAC) program. We have assembled interdisciplinary teams
and collaborations to develop the necessary state-of-the-art
mathematical algorithms and software, supported by appropriate hardware
and middleware infrastructure to use terascale computers effectively to
advance fundamental scientific research essential to the DOE mission.
These activities are central to the future of our mission. Advanced
scientific computing will continue to be a key contributor to
scientific research as we enter the twenty-first century. Major
scientific challenges exist in all Office of Science research programs
that can be addressed by advanced scientific supercomputing. Designing
materials atom-by-atom, revealing the functions of proteins,
understanding and controlling plasma turbulence, designing new particle
accelerators, and modeling global climate change, are just a few
examples.
* * *
Today, high-end scientific computation has reached a threshold
which we were all made keenly aware of when the Japanese Earth
Simulator was turned on. The Earth Simulator worked remarkably well on
real physical problems at sustained speeds that have never been
achieved before. The ability to get over 25 teraFLOPS in geophysical
science problems was not only an achievement, but it truly opened a new
world.
So the question before us at today's hearing--``Supercomputing: Is
the U.S. on the Right Path''--is very timely. There is general
recognition of the opportunities that high-end computation provides,
and this Administration has a path forward to meet this challenge.
The tools for scientific discovery have changed. Previously,
science had been limited to experiment and theory as the two pillars
for investigation of the laws of nature. With the advent of what many
refer to as ``Ultra-Scale'' computation,'' a third pillar--simulation--
has been added to the foundation of scientific discovery. Modern
computational methods are developing at such a rapid rate that
computational simulation is possible on a scale that is comparable in
importance with experiment and theory. The remarkable power of these
facilities is opening new vistas for science and technology.
Tradition has it that scientific discovery is based on experiment,
buttressed by theory. Sometimes the order is reversed, theory leads to
concepts that are tested and sometimes confirmed by experiment. But
more often, experiment provides evidence that drives theoretical
reasoning. Thus, Dr. Samuel Johnson, in his Preface to Shakespeare,
writes: ``Every cold empirick, when his heart is expanded by a
successful experiment, swells into a theorist.''
Many times, scientific discovery is counter-intuitive, running
against conventional wisdom. Probably the most vivid current example is
the experiment that demonstrated that the expansion of our Universe is
accelerating, rather than in steady state or contracting. We have yet
to understand the theoretical origins for this surprise.
During my scientific career, computers have developed from the now
``creaky'' IBM 701, upon which I did my thesis research, to the so-
called massively parallel processors or MPP machines, that fill rooms
the size of football fields, and use as much power as a small city.
The astonishing speeds of these machines, especially the Earth
Simulator, allow Ultra-Scale computation to inform our approach to
science, and I believe social sciences and the humanities. We are now
able to contemplate exploration of worlds never before accessible to
mankind. Previously, we used computers to solve sets of equations
representing physical laws too complicated to solve analytically. Now
we can simulate systems to discover physical laws for which there are
no known predictive equations. We can model physical or social
structures with hundreds of thousands, or maybe even millions, of
``actors,'' interacting with one another in a complex fashion. The
speed of our new computational environment allows us to test different
inter-actor (or inter-personal) relations to see what macroscopic
behaviors can ensue. Simulations can determine the nature of the
fundamental ``forces'' or interactions between ``actors.''
Computer simulation is now a major force for discovery in its own
right.
We have moved beyond using computers to solve very complicated sets
of equations to a new regime in which scientific simulation enables us
to obtain scientific results and to perform discovery in the same way
that experiment and theory have traditionally been used to accomplish
those ends. We must think of high-end computation as the third of the
three pillars that support scientific discovery, and indeed there are
areas where the only approach to a solution is through high-end
computation--and that has consequences.
* * *
American industry certainly is fully conversant with the past,
present and prospective benefits of high-end computation. The Office of
Science has received accolades for our research accomplishments from
corporations such as General Electric and General Motors. We have met
with the vice presidents for research of these and other member
companies of the Industrial Research Institute. We learned, for
example, that GE is using simulation very effectively to detect flaws
in jet engines. What's more, we were told that, if the engine flaws
identified by simulation were to go undetected, the life cycle of those
GE machines would be reduced by a factor of two--and that would cause
GE a loss of over $100,000,000.00.
The market for high-end computation extends beyond science, into
applications, creating a commercial market for ultra-scale computers.
The science and technology important to industry can generate
opportunities measured in hundreds of million, and perhaps billions of
dollars.
Here are just a few examples:
From General Motors:
``General Motors currently saves hundreds of millions of dollars by
using its in-house high performance computing capability of more than
3.5 teraFLOPS in several areas of its new vehicle design and
development processes. These include vehicle crash simulation, safety
models, vehicle aerodynamics, thermal and combustion analyses, and new
materials research. The savings are realized through reductions in the
costs of prototyping and materials used.
However, the growing need to meet higher safety standards, greater
fuel efficiency, and lighter but stronger materials, demands a steady
yearly growth rate of 30 to 50 percent in computational capabilities
but will not be met by existing architectures and technologies.. . .A
computing architecture and capability on the order of 100 teraFLOPS for
example would have quite an economic impact, on the order of billions
of dollars, in the commercial sector in its product design,
development, and marketing.''
And from General Electric:
``Our ability to model, analyze and validate complex systems is a
critical part of the creation of many of our products and design. Today
we make extensive use of high-performance computing based technologies
to design and develop products ranging from power systems and aircraft
engines to medical imaging equipment. Much of what we would like to
achieve with these predictive models is out of reach due to limitations
in current generation computing capabilities. Increasing the fidelity
of these models demands substantial increases in high-performance
computing system performance. We have a vital interest in seeing such
improvements in the enabling high-performance computing technologies..
. .In order to stay competitive in the global marketplace, it is of
vital importance that GE can leverage advances in high-performance
computing capability in the design of its product lines. Leadership in
high-performance computing technologies and enabling infrastructure is
vital to GE if we wish to maintain our technology leadership.''
Consider the comparison between simulations and prototyping for GE
jet engines.
For evaluation of a design alternative for the purpose of
optimization of a compressor for a jet engine design, GE would require
3.1 1018 floating point operations, or over a month at a
sustained speed of one teraFLOP, which is today's state-of-the-art. To
do this for the entire engine would require sustained computing power
of 50 teraFLOPS for the same period. This is to be compared with
millions of dollars, several years, and designs and re-designs for
physical prototyping.
Opportunities abound in other fields such as pharmaceuticals, oil
and gas exploration, and aircraft design.
The power of advance scientific computation is just beginning to be
realized. One reason that I have emphasized this so much is that some
seem to think that advanced scientific computation is the province of
the Office of Science and other federal science agencies and therefore
is not attractive to the vendors in this field. I believe that's
incorrect. I believe instead that our leading researchers are laying
out a direction and an understanding of available opportunities. These
opportunities spur markets for high-end computation quite comparable to
the commercial market which we have seen in the past but requiring the
efficiencies and the speeds which high-end computation can provide.
* * *
Discovery through simulation requires sustained speeds starting at
50 to 100 teraFLOPS to examine problems in accelerator science and
technology, astrophysics, biology, chemistry and catalysis, climate
prediction, combustion, computational fluid dynamics, computational
structural and systems biology, environmental molecular science, fusion
energy science, geosciences, groundwater protection, high energy
physics, materials science and nanoscience, nuclear physics, soot
formation and growth, and more (see http://www.ultrasim.info/
doe-docs/).
Physicists in Berkeley, California, trying to determine whether our
universe will continue to expand or eventually collapse, gather data
from dozens of distant supernovae. By analyzing the data and simulating
another 10,000 supernovae on supercomputers (at the National Energy
Research Scientific Computing Center or NERSC) the scientists conclude
that the universe is expanding--and at an accelerating rate.
I just returned from Vienna, where I was privileged to lead the
U.S. delegation in negotiations on the future direction for ITER, an
international collaboration that hopes to build a burning plasma fusion
reactor, which holds out promise for the realization of fusion power.
The United States pulled out of ITER in 1998. We're back in it this
year. What changed were simulations that showed that the new ITER
design will in fact be capable of achieving and sustaining burning
plasma. We haven't created a stable burning plasma yet, but the
simulations give us confidence that the experiments which we performed
at laboratory scales could be realized in larger machines at higher
temperatures and densities.
Looking to the future, we are beginning a Fusion Simulation Project
to build a computer model that will fully simulate a burning plasma to
both predict and interpret ITER performance and, eventually, assist in
the design of a commercially feasible fusion power reactor. Our best
estimate, however is that success in this effort will require at least
50 teraFLOPS of sustained computing power.
Advances in scientific computation are also vital to the success of
the Office of Science's Genomes to Life program.
The Genomes to Life program will develop new knowledge about how
micro-organisms grow and function and will marry this to a national
infrastructure in computational biology to build a fundamental
understanding of living systems. Ultimately this approach will offer
scientists insights into how to use or replicate microbiological
processes to benefit the Nation.
In particular, the thrust of the Genomes to Life program is aimed
directly at Department of Energy concerns: developing new sources of
energy; mitigating the long-term effects of climate change through
carbon sequestration; cleaning up the environment; and protecting
people from adverse effects of exposure to environmental toxins and
radiation.
All these benefits--and more--will be possible as long as the
Genomes to Life program achieves a basic understanding of thousands of
microbes and microbial systems in their native environments over the
next 10 to 20 years. To meet this challenge, however, we must address
huge gaps not only in knowledge but also in technology, computing, data
storage and manipulation, and systems-level integration.
The Office of Science also is a leader in research efforts to
capitalize on the promise of nanoscale science.
In an address to the American Association for the Advancement of
Science in February 2002, Dr. John Marburger, Director of the Office of
Science and Technology Policy, noted, ``. . .[W]e are in the early
stages of a revolution in science nearly as profound as the one that
occurred early in the last century with the birth of quantum
mechanics,'' a revolution spurred in part by ``the availability of
powerful computing and information technology.''
``The atom-by-atom understanding of functional matter,'' Dr.
Marburger continued, ``requires not only exquisite instrumentation, but
also the capacity to capture, store and manipulate vast amounts of
data. The result is an unprecedented ability to design and construct
new materials with properties that are not found in nature.. . .[W]e
are now beginning to unravel the structures of life, atom-by-atom using
sensitive machinery under the capacious purview of powerful
computing.''
In both nanotechnology and biotechnology, this revolution in
science promises a revolution in industry. In order to exploit that
promise, however, we will need both new instruments and more powerful
computers, and the Office of Science has instituted initiatives to
develop both.
We have begun construction at Oak Ridge National Laboratory on the
first of five Nanoscale Science Research Centers located to take
advantage of the complementary capabilities of other large scientific
facilities, such as the Spallation Neutron Source at Oak Ridge, our
synchrotron light sources at Argonne, Brookhaven and Lawrence Berkeley,
and semiconductor, microelectronics and combustion research facilities
at Sandia and Los Alamos. When complete, these five Office of Science
nanocenters will provide the Nation with resources unmatched anywhere
else in the world.
To determine the level of computing resources that will be
required, the Office of Science sponsored a scientific workshop on
Theory and Modeling in Nanoscience, which found that simulation will be
critical to progress, and that new computer resources are required. As
a first step to meeting that need, our Next Generation Architecture
initiative is evaluating different computer architectures to determine
which are most effective for specific scientific applications,
including nanoscience simulations.
There are many other examples where high-end computation has
changed and will change the nature of the field. My own field is
complex systems. I work in a somewhat arcane area called spin glasses,
where we can examine the dynamic properties of these very complex
systems, which in fact are related to a number of very practical
applications. Through scientific simulation, a correlation length was
predicted for a completely random material, a concept unknown before.
Simulation led to the discovery that there was a definable correlation
length in this completely random system. Our experiments confirmed this
hypothesis. Again, insights were created that simply were not possible
from a set of physical equations that needed solutions, with observable
consequences. There are countless opportunities and examples where
similar advances could be made.
* * *
As the Chairman and Members of this committee know, the Bush
Administration shares Congress' keen interest in high-end computation
for both scientific discovery and economic development. A senior
official of the Office of Science is co-chairing the interagency High-
End Computing Revitalization Task Force, which includes representatives
from across the government and the private sector. We are working very
closely with the NNSA, the Department of Defense, the National
Aeronautics and Space Administration, the National Science Foundation,
and the National Institutes of Health to assess how best to coordinate
and leverage our agencies' high-end computation investments now and in
the future.
DOE is playing a major role in the task force through the Office of
Science and the NNSA, and many of the task force's findings and plans
are based on Office of Science practices in advanced computing and
simulation.
One of the major challenges in this area is one of metrics. How do
we know what we are trying to accomplish, and how can we measure how
we're getting there? What are the opportunities? What are the barriers?
What should we be addressing as we begin to explore this new world?
Our problem in the past has been that, where we have large
computational facilities, we have cut them up in little pieces and the
large-scale scientific programs that some researchers are interested in
have never really had a chance to develop. There's nothing wrong with
our process; it is driven by a peer review system. But for some
promising research efforts, there simply have not been enough cycles or
there wasn't an infrastructure which would allow large-scale
simulations to truly develop and produce the kind of discoveries we
hope to achieve.
Recognizing this, the Office of Science has announced that ten
percent of our National Energy Research Scientific Computing Center at
Lawrence Berkeley National Laboratory--now at ten teraFLOP peak speed--
is going to be made available for grand challenge calculations. We are
literally going to carve out 4.5 million processor hours and 100
terabytes of disk space for perhaps four or five scientific problems of
major importance. We are calling this initiative INCITE--the Innovative
and Novel Computational Impact on Theory and Experiment--and we expect
to be ready to proceed with it around August 1, 2003. At that time, we
will open the competition to all, whether or not they are affiliated
with or funded by DOE.
We are launching the INCITE initiative for two reasons. For one,
it's the right thing to do: there are opportunities for major
accomplishments in this field of science. In addition, there is also a
``sociology'' that we need to develop.
Given the size and complexity of the machines required for
sustained speeds in the 50 to 100 teraFLOPS regime, the sociology of
high-end computation will probably have to change. One can think of the
usage of ultra-scale computers as akin to that of our current light
sources: large machines used by groups of users on a shared basis.
Following the leadership of our SciDAC program, interdisciplinary teams
and collaborators will develop the necessary state-of-the-art
mathematical algorithms and software, supported by appropriate hardware
and middleware infrastructure, to use terascale computers effectively
to advance fundamental research in science. These teams will associate
on the basis of the mathematical infrastructure of problems of mutual
interest, working with efficient, balanced computational architectures.
The large amount of data, the high sustained speeds, and the cost
will probably lead to concentration of computing power in only a few
sites, with networking useful for communication and data processing,
but not for core computation at terascale speeds. Peer review of
proposals will be used to allocate machine time. Industry will be
welcome to participate, as has happened in our light sources. Teams
will make use of the facilities as user groups, using significant
portions (or all) of the machine, depending on the nature of their
computational requirements. Large blocks of time will enable scientific
discovery of major magnitude, justifying the large investment ultra-
scale computation will require.
We will open our computational facilities to everyone. Ten percent
of NERSC's capability will be available to the entire world.
Prospective users will not have to have a DOE contract, or grant, or
connection. The applications will be peer reviewed, and will be judged
solely on their scientific merit. We need to learn how to develop the
sociology that can encourage and then support computation of this
magnitude; this is a lot of computer time. It may be the case that
teams rather than individuals will be involved. It even is possible
that one research proposal will be so compelling that the entire ten
percent of NERSC will be allocated to that one research question.
The network that may be required to handle that amount of data has
to be developed. There is an ES network which we are involved in, and
we are studying whether or not it will be able to handle the massive
amounts of data that could be produced under this program.
We need to get scientific teams--the people who are involved in
algorithms, the computer scientists, and the mathematicians--together
to make the most efficient use of these facilities. That's what this
opening up at NERSC is meant to do. We want to develop the community of
researchers within the United States--and frankly around the world--
that can take advantage of these machines and produce the results that
will invigorate and revolutionize their fields of study.
But this is just the beginning.
* * *
As we develop the future high-end computational facilities for this
nation and world, it is clearly our charge and our responsibility to
develop scientific opportunities for everyone. This has been the U.S.
tradition. It has certainly been an Office of Science tradition, and we
intend to see that this tradition continues, and not just in the
physical sciences.
We are now seeing other fields recognizing that opportunities are
available to them. In biology, we are aware that protein folding is a
very difficult but crucial issue for cellular function. The time scales
that biologists work with can scale from a femto-second to seconds-a
huge span of time which our current simulation capabilities are unable
to accommodate.
High-performance computing provides a new window for researchers to
observe the natural world with a fidelity that could only be imagined a
few years ago. Research investments in advanced scientific computing
will equip researchers with premier computational tools to advance
knowledge and to solve the most challenging scientific problems facing
the Nation.
With vital support from this committee, the Congress and the
Administration, we in the Office of Science will help lead the U.S.
further into the new world of supercomputing.
We are truly talking about scientific discovery. We are talking
about a third pillar of support. We are talking about opportunities to
understand properties of nature that have never before been explored.
That's the concept, and it explains the enormous excitement that we
feel about this most promising field.
We are very gratified that this committee is so intent on enabling
us to pursue so many important opportunities, for the sake of
scientific discovery, technological innovation, and economic
competitiveness.
Thank you very much.
Biography for Raymond L. Orbach
Dr. Raymond L. Orbach was sworn in as the 14th Director of the
Office of Science at the Department of Energy (DOE) on March 14, 2002.
As Director of the Office of Science (SC), Dr. Orbach manages an
organization that is the third largest federal sponsor of basic
research in the United States and is viewed as one of the premier
science organizations in the world. The SC fiscal year 2002 budget of
$3.3 billion funds programs in high energy and nuclear physics, basic
energy sciences, magnetic fusion energy, biological and environmental
research, and computational science. SC, formerly the Office of Energy
Research, also provides management oversight of the Chicago and Oak
Ridge Operations Offices, the Berkeley and Stanford Site Offices, and
10 DOE non-weapons laboratories.
Prior to his appointment, Dr. Orbach served as Chancellor of the
University of California (UC), Riverside from April 1992 through March
2002; he now holds the title Chancellor Emeritus. During his tenure as
Chancellor, UC-Riverside grew from the smallest to one of the most
rapidly growing campuses in the UC system. Enrollment increased from
8,805 to more than 14,400 students with corresponding growth in faculty
and new teaching, research, and office facilities.
In addition to his administrative duties at UC-Riverside, Dr.
Orbach maintained a strong commitment to teaching. He sustained an
active research program; worked with postdoctoral, graduate, and
undergraduate students in his laboratory; and taught the freshman
physics course each winter quarter. As Distinguished Professor of
Physics, Dr. Orbach set the highest standards for academic excellence.
From his arrival, UC-Riverside scholars led the Nation for seven
consecutive years in the number of fellows elected to the prestigious
American Association for the Advancement of Science (AAAS).
Dr. Orbach began his academic career as a postdoctoral fellow at
Oxford University in 1960 and became an assistant professor of applied
physics at Harvard University in 1961. He joined the faculty of the
University of California, Los Angeles (UCLA) two years later as an
associate professor, and became a full professor in 1966. From 1982 to
1992, he served as the Provost of the College of Letters and Science at
UCLA.
Dr. Orbach's research in theoretical and experimental physics has
resulted in the publication of more than 240 scientific articles. He
has received numerous honors as a scholar including two Alfred P. Sloan
Foundation Fellowships, a National Science Foundation Senior
Postdoctoral Fellowship, a John Simon Guggenheim Memorial Foundation
Fellowship, the Joliot Curie Professorship at the Ecole Superieure de
Physique et Chimie Industrielle de la Ville de Paris, the Lorentz
Professorship at the University of Leiden in the Netherlands, and the
1991-1992 Andrew Lawson Memorial Lecturer at UC-Riverside. He is a
fellow of the American Physical Society and the AAAS.
Dr. Orbach has also held numerous visiting professorships at
universities around the world. These include the Catholic University of
Leuven in Belgium, Tel Aviv University, and the Imperial College of
Science and Technology in London. He also serves as a member of 20
scientific, professional, or civic boards.
Dr. Orbach received his Bachelor of Science degree in Physics from
the California Institute of Technology in 1956. He received his Ph.D.
degree in Physics from the University of California, Berkeley, in 1960
and was elected to Phi Beta Kappa.
Dr. Orbach was born in Los Angeles, California. He is married to
Eva S. Orbach. They have three children and seven grandchildren.
Chairman Boehlert. Thank you very much, Dr. Orbach.
You noticed the red light was on, and the Chair was
generous with the time. As I said, I am not going to be
arbitrary. I wish the appropriators were as generous with the
funding for your office as we are with time for your views, but
we are working continually together on that one.
Dr. Freeman.
STATEMENT OF DR. PETER A. FREEMAN, ASSISTANT DIRECTOR, COMPUTER
AND INFORMATION SCIENCE AND ENGINEERING DIRECTORATE, NATIONAL
SCIENCE FOUNDATION
Dr. Freeman. Good morning, Mr. Chairman, Mr. Hall, and
distinguished Members of the Committee. NSF deeply appreciates
this committee's long time support and recognizes your special
interest in computing, so I am delighted to be here today to
discuss those topics with you.
Supercomputing is a field that NSF, as you noted in your
opening remarks, has championed for many years. And it is one
in which we intend to continue to lead the way. At the same
time, we are committed to realizing the compelling vision
described in the report of the recent NSF Advisory Panel on
Cyberinfrastructure, commonly known as the Atkins Committee.
They have forcefully told us, and I quote, ``A new age has
dawned in scientific and engineering research, '' that will be
enabled by cyberinfrastructure.
The term ``cyberinfrastructure'' sounds exotic and is
sometimes confused as being something new, but in reality, it
is intended to signify a set of integrated facilities and
services essential to the conduct of leading edge science and
engineering. Cyberinfrastructure must include a range of
supercomputers as well as massive storage, high performance
networks, databases, lots of software, and above all, highly
trained people. An advanced cyberinfrastructure, with
supercomputing as an important element, promises to
revolutionize research in the 21st Century. The opportunities
that are presented to us in this area must be exploited for the
benefit of all of our citizens for their continuing health,
security, education, and wealth.
We are committed to a key recommendation of the Atkins
Report, namely that NSF, in partnership with other agencies and
other organizations, must make significant investments in the
creation, deployment, and application of advanced
cyberinfrastructure to empower continued U.S. leadership in
science and engineering.
To bring about this scientific revolution, we must maintain
a broad discourse about cyberinfrastructure. Supercomputers are
essential, but without software, networks, massive storage,
databases, and trained people all integrated securely, they
will not deliver their potential. Further, there are now many
areas of science that need this integrated and balanced support
or balanced approach more than they need any single element.
My written testimony provides detailed answers to the
questions you posed for this hearing, but allow me to summarize
them. NSF will most definitely continue its commitment to
supercomputing and will provide the support necessary to
utilize it in all of science and engineering. Supercomputing
capability is an essential component in the cyberinfrastructure
vision, and we are committed to expanding this capability in
the future.
NSF's recent investments in grid computing underscore the
importance of integrating supercomputing within a
cyberinfrastructure. Indeed, the first increment of our funding
and our terascale efforts was for a supercomputer, which at the
time it was installed, was the most powerful open access
machine in the research world. This machine will be one of the
main resources on the grid that is currently under
construction.
The Atkins Report recommended ``a 2-year extension of the
current PACI [Partnerships for Advanced Computational
Infrastructure] cooperative agreements,'' and we have done
that. The panel also recommended ``new separately peer-reviewed
enabling and application infrastructure'' programs. We have
identified funding for these efforts and are in the process of
working out the details for these new programs. Finally, the
Atkins Report recommends that ``until the end of the original
10-year lifetime of the PACI program,'' the centers ``should
continue to be assured of stable, protected funding to provide
the highest-end computing resources.'' NSF is currently
gathering community input on how best to structure future
management of these resources, and will be announcing our
specific plans in the coming months.
With regard to coordination, NSF has been an active
participant in all four of the subgroups of the OSTP planning
activity for high-end computing. We are coordinating closely at
all levels with OSTP and our sister agencies to make sure that
the recommendations of the task force are carried forward in an
effective manner.
NSF does not see a divergence in the needs of industry and
those of the research community. As researchers push the
boundaries in their work, the results are often quickly picked
up by industry. In other areas, industry has had to tackle
problems first, such as data mining, and now those techniques
are becoming useful in scientific research. This symbiotic
relationship has worked very well in the past, and we intend to
continue it in the future.
Mr. Chairman, let me close by providing an answer to the
overall question of whether we are on the right path in
supercomputing. My answer is yes; if we keep in perspective
that supercomputers must be embedded in a cyberinfrastructure
that also includes massive storage, high-performance networks,
databases, lots of software, well-trained people, and that the
entire ensemble be open to all scientists and engineers.
Thank you, and I will be glad to answer any questions you
may have.
[The prepared statement of Dr. Freeman follows:]
Prepared Statement of Peter A. Freeman
Good morning, Mr. Chairman and Members of the Committee. I am Dr.
Peter Freeman, Assistant Director of the NSF for CISE.
Introduction
I am delighted to have the opportunity to testify before you this
morning and to discuss the topic Supercomputing: Is the U.S. on the
Right Path? Supercomputers are an extremely important national resource
in many sectors of our society, and for decades, these resources have
yielded astounding scientific breakthroughs. Supercomputing is a field
that NSF has championed and supported for many years and it is one in
which we will continue to lead the way.
There seems to be some confusion in the scientific community as to
NSF's commitment to High-End Computing (HEC) which is the current term
being used for ``supercomputing.'' I want to clear the air this
morning. Before I briefly summarize my written testimony, let me state
unequivocally that NSF remains absolutely committed to providing
researchers the most advanced computing equipment available and to
sponsoring research that will help create future generations of
computational infrastructure, including supercomputers.
At the same time, we are committed to realizing the compelling
vision described in the report of the NSF Advisory Panel on
Cyberinfrastructure, commonly known as the Atkins Committee--that ``a
new age has dawned in scientific and engineering research, pushed by
continuing progress in computing, information and communications
technology.'' This cyberinfrastructure includes, and I quote, ``not
only high-performance computational services, but also integrated
services for knowledge management, observation and measurement,
visualization and collaboration.''
The scientific opportunities that lie before us in many fields can
only be realized with such a cyberinfrastructure. Just as
supercomputing promised to revolutionize the conduct of science and
engineering research several decades ago, and we are seeing the results
of that promise today, so does an advanced cyberinfrastructure promise
to revolutionize the conduct of science and engineering research and
education in the 21st century. The opportunities that a balanced,
state-of-the-art cyberinfrastructure promises must be exploited for the
benefit of all of our citizens--for their continuing health, security,
education, and wealth.
To be clear, we are committed to what the Atkins Report and many
others in the community, both formally and informally, are saying: That
NSF, in partnership with other public and private organizations, must
make investments in the creation, deployment and application of
cyberinfrastructure in ways that radically empower all science and
engineering research and allied education. . .thereby empowering what
the Atkins Report defines as ``a revolution.''
Cyberinfrastructure--with HEC as an essential component--can bring
about this true revolution in science and engineering. It promises
great advances for all of the areas of our society served by science
and engineering, but it will be realized only if we stay focused on the
value of all components of cyberinfrastructure.
Supercomputers have been one of the main drivers in this revolution
up to now because of the continuing evolution of computing technology.
Computers were initially developed to deal with pressing, numerical
computations and they will continue to be extremely important. In
recent years, however, many scientific advances have been enabled not
only by computers, but by the great expansion in the capacity of
computer storage devices and communication networks, coupled now with
rapidly improving sensors.
There are now many examples of revolutionary scientific advances
that can only be brought about by utilizing other components of
cyberinfrastructure in combination with HEC. This necessary convergence
means that we must maintain a broad discourse about
cyberinfrastructure. As we set about building and deploying an advanced
cyberinfrastructure, we will ensure that the HEC portion remains an
extremely important component in it.
History of NSF Support for Supercomputing
NSF has been in the business of supporting high performance
computation in the form of centers since the establishment of the first
Academic Computing Centers in the 1960's. As computers became
increasingly powerful, they were later designated to be
``supercomputers.'' In the mid-1980's, NSF created the first
supercomputing centers for the open science community. A decade later,
support was established through the Partnerships for Advanced
Computational Infrastructure (PACI) program. In 2000, a parallel
activity that is now known as the Extensible Terascale Facility was
initiated. (There is much more to this history that is available if
desired.) Beginning in FY 2005, support will be provided through
cyberinfrastructure program(s) currently under development.
Over time, technological innovations have led to movement away from
the use of the term ``supercomputing centers'' since it inadequately
describes the full measure and promise of what is being done at such
centers, and what is at stake. The idea of the ``supercomputer'' lives
on as a legacy, however a more accurate title for this kind of
infrastructure would be High-performance Information Technology (HIT)
Centers.
NSF currently supports three major HIT Centers: the San Diego
Supercomputer Center (SDSC), the National Center for Supercomputing
Applications (NCSA), and the Pittsburgh Supercomputer Center (PSC).
These centers have been participating in this enterprise since the days
of the first supercomputer centers in the early 1980's. They have
evolved steadily over the past quarter century and now represent
special centers of talent and capability for the science community
broadly and the Nation at large.
In the last six years, NSF's support for these HIT Centers has been
provided predominantly through the PACI program. More recently, a new
consortium of HIT Centers has emerged around the new grid-enabled
concept of the Extensible Terascale Facility (ETF).
Current Activities
In order to describe NSF's current activities in the area of
supercomputing, I'd like to respond directly to the following questions
formulated by Chairman Boehlert and the Committee on Science. (Note:
Italicized and numbered statements are drawn verbatim from Chairman
Boehlert's letter of invitation.)
1. Some researchers within the computer science community have
suggested that the NSF may be reducing its commitment to the
supercomputer centers. Is this the case?
NSF is most definitely not reducing its commitment to
supercomputing.
For several decades the agency has invested millions of taxpayer
dollars in the development and deployment of a high-end computational
infrastructure. These resources are made widely available to the
science and engineering research and education community. The agency is
not reducing its commitment to such efforts. In fact, leading-edge
supercomputing capabilities are an essential component in the
cyberinfrastructure and, in line with the recommendations of the
Advisory Panel on Cyberinfrastructure, we are committed to expanding
such capabilities in the future.
1. (cont'd.) To what extent does the focus on grid computing represent
a move away from providing researchers with access to the most advanced
computing equipment?
The term ``grid computing'' is ambiguous and often misused. It
sometimes is used to signify a single computational facility composed
of widely separated elements (nodes) that are interconnected by high-
performance networks and that are operating in a manner that the user
sees a single ``computer.'' This is a special case of the more general
concept of a set of widely separated computational resources of
different types, which can be accessed and utilized as their particular
capabilities are needed.
While still experimental at this stage, grid computing promises to
become the dominant modality of High-performance IT (and, eventually,
of commodity computing). One need only think about the World Wide Web
(WWW) to understand the compelling importance of grid computing. In the
WWW one is able from a single terminal (typically a desk-top PC) to
access many different databases, on-line services, even computational
engines today. Imagine now that the nodes that are accessible in this
way are HECs with all of their computational power; or massive data
stores that can be manipulated and analyzed; or sophisticated
scientific instruments to acquire data; or any of a dozen other
foreseen and unforeseen tools. With this vision, perhaps one can
understand the promise of grid computing.
NSF's recent investments in grid computing, through the ETF, should
not be seen as a reduction in the agency's commitment to HEC. Rather,
it underscores the importance of HEC integrated into a broad
cyberinfrastructure. Indeed, the first increment of ETF funding was for
a HEC machine, which at the time it was installed at the PSC, was the
most powerful open access machine in the research world (and it is,
three years later, still number 9 on the Top 500 list). This machine is
one of the main resources on the ETF grid. While NSF may not always
have the world's fastest machine, we will continue to provide a range
of supercomputing systems that serve the ever increasing and changing
needs of science and engineering.
At the same time, the ETF investment in grid computing is not the
ONLY investment the agency has recently made in HEC. In fact, HEC
upgrades at NCSA and SDSC during FY 2003 and 2004 are expected to fund
the acquisition of an additional 20 Teraflops of HEC capability.
NSF's unwavering commitment is to continuously advance the
frontier, and grid computing is widely acknowledged to represent the
next frontier in computing. In short, the ETF represents our commitment
to innovation at the computing frontier.
2. What are the National Science Foundation's (NSF's) plans for
funding the supercomputer centers beyond fiscal year 2004? To what
extent will you be guided by the recommendation of the NSF Advisory
Panel on Cyberinfrastructure to maintain the Partnerships for Advanced
Computational Infrastructure, which currently support the supercomputer
centers?
NSF's plans for funding supercomputer centers beyond FY 2004 are
very much guided by the recommendations of the NSF Advisory Panel on
Cyberinfrastructure as described below.
In its report, the Panel recommended ``a two-year
extension of the current PACI co-operative agreements.'' The
National Science Board approved the second one-year extension
of the PACI cooperative agreements at the May 2003 meeting.
The Panel also recommended that ``. . .the new
separately peer-reviewed enabling and application
infrastructure would begin in 2004 or 2005, after the two-year
extensions of the current cooperative agreements.''
The President requested $20 million for NSF in FY 2004 for
activities that will focus on the development of
cyberinfrastructure, including enabling infrastructure (also
known as enabling technology). This increased investment in
enabling technology will strengthen the agency's portfolio of
existing awards, and as the Panel recommended, awards will be
identified through merit-review competition.
In addition, NSF will also increase its investments in
applications infrastructure (also known as applications
technology) by drawing upon interdisciplinary funds available
in the FY 2004 ITR priority area activity. Again, the most
promising proposals will be identified using NSF's rigorous
merit review process.
Finally, the Panel's Report recommends that ``After
these two years, until the end of the original 10-year lifetime
of the PACI program, the panel believes that'' NCSA, SDSC and
PSC ``should continue to be assured of stable, protected
funding to provide the highest-end computing resources.''
Accordingly, and in planning how to provide such support
while positioning the community to realize the promise of
cyberinfrastructure, NSF has held a series of workshops and
town hall meetings over the course of the past two months to
gather community input. Informed by community input, support
for SDSC, NCSA and PSC will be provided through new awards to
be made effective the beginning of FY 2005.
NSF has also committed to providing support for the management and
operations of the Extensible Terascale Facility through FY 2009; this
includes support for SDSC, NCSA and PSC who are partners in the ETF.
3. To what extent will NSF be guided by the recommendations of the
High-End Computing Revitalization Task Force? How will NSF contribute
to the Office of Science and Technology Policy plan to revitalize high-
end computing?
NSF has been an active participant in all four of the subgroups of
the OSTP current planning activity called HEC-RTF. The final report is
not finished, but we continue to coordinate closely at all levels with
OSTP and its sister agencies to make sure that the recommendations of
the Task Force are carried forward in a productive and effective
manner. NSF's traditional role as a leader in innovating new HEC
computational mechanisms and applications, and in ensuring that there
are appropriate educational programs in place to train scientists and
engineers to use them, must be integral to any efforts to increase HEC
capabilities for the Nation.
We are now planning our request for the FY 2005 budget, and
cyberinfrastructure, including HEC, is likely to be a major component
in it. We intend to continue to have more than one high-end machine for
the NSF community to access and to invest in needed HEC capabilities as
noted above.
4. To what extent are the advanced computational needs of the
scientific community and of the private sector diverging? What is the
impact of any such divergence on the advanced computing programs at
NSF?
I don't believe that the advanced computational ``needs'' of the
science community and the private sector are diverging. In fact, I
believe that the growing scientific use of massive amounts of data
parallels what some sectors of industry already know. In terms of HEC,
it is clear that both communities need significantly faster machines to
address those problems that can only be solved by massive computations.
For several decades, NSF has encouraged its academic partners in
supercomputing, including NCSA, SDSC and PSC, to develop strong
relationships with industry. The initial emphasis was on
``supercomputing.'' And many of the industrial partners at the centers
learned about supercomputing in this way, and then started their own
internal supercomputer centers.
When Mosaic (the precursor to Netscape) was developed at NCSA,
industry was able to rapidly learn about and exploit this new,
revolutionary technology. As noted above, grid computing, which is
being innovated at NCSA, SDSC and PSC and their partners today, is
already being picked up by industry as a promising approach to
tomorrow's computing problems.
As researchers push the boundaries in their work, the results (and
means of obtaining those results) are often quickly picked up by
industry. Conversely, in some areas industry has had to tackle problems
(such as data mining) first and now those techniques are becoming
useful in scientific research. We intend to continue the close
collaboration that has existed for many years.
Conclusion
Mr. Chairman, I hope that this testimony dispels any doubt about
NSF's commitment to HEC and the HIT Centers that today provide
significant value to the science and engineering community.
I hope that I have also been able to articulate that the
cyberinfrastructure vision eloquently described in the Atkins report
includes HEC and other advanced IT components. This cyberinfrastructure
will enable a true revolution in science and engineering research and
education that can bring unimagined benefits to our society.
NSF recognizes the importance of HEC to the advanced scientific
computing infrastructure for the advancement of science and knowledge.
We are committed to continuing investments in HEC and to developing new
resources that will ensure that the United States maintains the best
advanced computing facilities in the world. We look forward to working
with you to ensure that these goals are fulfilled.
Thank you for the opportunity to appear before you this morning.
Biography for Peter A. Freeman
Peter A. Freeman became Assistant Director for the Computer and
Information Science and Engineering Directorate (CISE) on May 6, 2002.
Dr. Freeman was previously at Georgia Institute of Technology as
professor and founding Dean of the College of Computing since 1990. He
served in that capacity as the John P. Imlay, Jr. Dean of Computing,
holding the first endowed Dean's Chair at Georgia Tech. He also served
as CIO for the campus for three years. In addition, as a general
officer of the campus, he was heavily involved in planning and
implementing a wide range of activities for the campus including a
successful $700M capital campaign and the Yamacraw Economic Development
Mission. He was in charge of the FutureNet Project, part of the campus
technology preparations for the 1996 Olympic Village, that resulted in
a very high-performance and broad campus network. In 1998, he chaired
the Sam Nunn NationsBank Policy Forum on Information Security which
lead to the creation of the Georgia Tech Information Security Center,
one of the first comprehensive centers in the country focused on
information security.
During 1989-90 Dr. Freeman was Visiting Distinguished Professor of
Information Technology at George Mason University in Fairfax, Virginia,
and from 1987 to 1989 he served as Division Director for Computer and
Computation Research at the National Science Foundation. He served on
the faculty of the Department of Information and Computer Science at
the University of California, Irvine, for almost twenty years before
coming to Georgia Tech.
He co-authored The Supply of Information Technology Workers in the
United States (CRA, 1999) and authored Software Perspectives: The
System is the Message (Addison Wesley, 1987), Software Systems
Principles (SRA, 1975), and numerous technical papers. In addition, he
edited or co-edited four books including, Software Reusability (IEEE
Computer Society, 1987), and Software Design Techniques, 4th edition
(IEEE Press, 1983). He was the founding editor of the McGraw-Hill
Series in Software Engineering and Technology, has served on several
editorial boards and numerous program committees, and was an active
consultant to industry, academia, and government.
Dr. Freeman was a member of the Board of Directors of the Computing
Research Association (1988-2002), serving as Vice-Chair and Chair of
the Government Affairs Committee. He was a member of select review
committees of the IRS and FAA Airtraffic Control modernization efforts,
and has served on a variety of national and regional committees. While
at NSF, he helped formulate the High-Performance Computing and
Communications Initiative of the Federal Government.
Dr. Freeman is a Fellow of the IEEE (Institute for Electrical and
Electronics Engineers), AAAS (American Association for the Advancement
of Science), and the ACM (Association for Computing Machinery). He
received his Ph.D. in computer science from Carnegie-Mellon University
in 1970, his M.A. in mathematics and psychology from the University of
Texas at Austin in 1965, and his B.S. in physics from Rice University
in 1963. His research and technical expertise has focused on software
systems and their creation. His earliest work (1961-63) involved
developing advanced scientific applications in the days before there
were operating systems and other support software. This led him to
design and build one of the earliest interactive time-sharing operating
systems (1964) and ultimately to early work applying artificial
intelligence to the design process for software (1965-75). This
culminated with the publication of his first book, Software System
Principles (SRA, 1975).
After a short stint teaching overseas for the United Nations, he
focused his work on software engineering, ultimately being recognized
for this early work by being elected a Fellow of the IEEE. Along with
Prof. A.I. Wasserman, he developed one of the first software design
courses (taken by thousands of industry practitioners) and published a
highly popular text that served as a first introduction to software
engineering. His research during this period focused on reusable
software, especially using formal transformation systems. That work has
resulted in several startup companies.
Since 1987 when he was ``loaned'' by the University of California
to the National Science Foundation, he has focused his attention on
national policy and local action intended to advance the field of
computing. In addition to his many activities as Dean at Georgia Tech,
he headed an NSF-funded national study of the IT worker shortage
(http://www.cra.org/reports/wits/cra.wits.html), started an active
group for Deans of IT& Computing, and published several papers relative
to future directions of the field.
Chairman Boehlert. Thank you very much, Dr. Freeman.
The Chair recognizes Mr. Johnson of Illinois.
Mr. Johnson. Thank you, Mr. Chairman. I appreciate the
opportunity to introduce our next guest, and I certainly
commend the Chairman and the Committee on the amble of the
hearing as well as the quality of the witnesses that we have
here.
It gives me a tremendous amount of pleasure to represent--
or to introduce to you Dr. Daniel A. Reed, who is the Director
of the National Center for Computing--Supercomputing
Applications at the University of Illinois, my home university,
the University of Illinois at Urbana-Champaign. Dr. Reed, who
is obviously here with us today, serves as the Director of the
National Computational Science Alliance, the National Center
for Computing--Supercomputing Applications at the University of
Illinois at Urbana-Champaign. Dr. Reed is one of two principal
investigators and the chief architect for the NSF TeraGrid
project to create a U.S. national infrastructure for grid
computing. He is an incoming member of the President's
Information Technology Advisory Committee and was formerly,
from 1996 to 2001, I believe, Dr. Reed was the head of the
University of Illinois Computer Science Department and a co-
leader of the National Computational Science Alliance Enabling
Technology team for three years.
He brings a tremendous amount of pride to our tremendous
university, which I think is recognized worldwide as one of the
leaders in this area. And the impact that Dr. Reed and our
center has had on the computer science field, not only here but
around the country and the world, for that matter, is really
indescribable.
So with those brief introductory remarks and my
appreciation to the Chair and the Committee as well as Dr.
Reed, it gives me a great deal of pleasure to introduce to the
Committee and--for his testimony, Dr. Daniel A. Reed.
Dr. Reed. Thank you, Congressman Johnson.
Chairman Boehlert. Dr. Reed, with a glowing introduction.
STATEMENT OF DR. DANIEL A. REED, DIRECTOR, NATIONAL CENTER FOR
SUPERCOMPUTING APPLICATIONS, UNIVERSITY OF ILLINOIS AT URBANA-
CHAMPAIGN
Dr. Reed. Good morning, Mr. Chairman, and Members of the
Committee. In response to your questions, I would like to make
three points today regarding the status and future of advanced
scientific computing.
First, the National Science Foundation has played a
critical role in providing the high-end computational
infrastructure for the science and engineering community, a
role that really must continue. Via this program, the NSF
Supercomputing Centers have provided community access to high-
end computers that, as you noted earlier, had previously been
available only in restricted cases and to researchers at
National Laboratories. This access is by national peer review,
which ensures that the most promising proposals from across
science and engineering benefit from high-end computing
support.
NSF's planned cyberinfrastructure will play a critical role
in accelerating scientific discovery, as Dr. Freeman noted, by
coupling scientific data archives, instruments, and computing
resources via distributed grids. The important issue, though,
and the one we are discussing today, is a relative level of
investment in distributed cyberinfrastructure and high-end
computing.
Grids, which were pioneered by the high-end computing
community, are not a substitute for high-end computing. Many
problems of national importance can only be solved by tightly
coupled high-end computing systems.
This brings me to my second point: the challenges before
us. On behalf of the Interagency High-End Computing
Revitalization Task Force, I recently chaired a workshop on
high-end computing opportunities and needs. And at the
workshop, researchers made compelling cases for sustained
performance levels of 25 to 100 teraflops to enable new
scientific discoveries. By comparison, the aggregate peak
performance of the system's now available in NSF's three
Supercomputing Centers and the Department of Energy's National
Energy Research Scientific Computing Center, or NERSC, is
roughly 25 teraflops. Simply put, there are neither enough
high-end computing systems available nor are there capabilities
adequate to address critical research opportunities. And
planned procurements, the ones on the horizon, will not fully
address that shortfall.
At the workshop, researchers also cited the difficulty in
achieving high performance on these machines. In the 1990's, as
it was mentioned earlier, the U.S. High-Performance Computing
Communications program supported the development of several new
machines. But in retrospect, we did not learn a critical
lesson, a lesson from the 1970's with vector systems, namely
the need for long-term, sustained investment in both hardware
and software. In the 1990's, we under-invested in software, and
we expected innovative research projects to yield robust,
mature software for use by the science community in only two to
three years.
In addition to that, we are now extremely dependent on
hardware derived, in my opinion, too narrowly from the
commercial market. Many, though I emphasize not all, scientific
applications and, indeed, several critical to weapons design,
signals intelligence, and cryptanalysis are not well matched to
the capabilities of commercial systems. I believe new high-end
computing designs will be needed to address some of these
scientific applications, both fully custom high-end designs, as
well as more appropriate designs based on commodity components.
This leads me to my third and final point: appropriate
levels of cooperation and support for high-end computing. I
believe we must change the model of development and deployment
for high-end systems in the U.S. if we are to sustain the
leadership needed for continued scientific discovery and
national security. The Japanese Earth System Simulator is a
wake-up call as it highlights the critical importance of
industry/government collaboration and sustained investment.
To sustain U.S. leadership, I believe we must pursue two
concurrent paths. First, we must continue to acquire and deploy
high-end systems, but at larger scale if we are to satisfy the
unmet demands of the open research community. As many have
noted, many community reports and panels have made this point,
including the recent NSF cyberinfrastructure report. NSF should
build on its high-end computing successes and implement the
high-end computing recommendations of the cyberinfrastructure
report, increasing and sustaining the long-term investment in
high-end computing hardware, software, and the support staff
needed to catalyze scientific discovery.
But the need for high-end computing is both broad and deep.
It has both research components as well as mission
applications. Hence, high-end computing system deployment
should not be viewed as an interagency competition, but rather
as an unmet need that requires aggressive responses from many
agencies. Their complimentary roles in each agency has a place
to play there.
Second, and concurrently, we must begin a coordinated R&D
effort to create systems that are better matched to scientific
application needs. I believe we must fund the construction of
large-scale prototypes with balanced exploration of new
hardware and software models driven critically by scientific
application requirements. This cycle of coordinated prototyping
assessment and commercialization must be a long-term and
sustaining investment. It cannot be a one-time crash program.
Thank you very much for your time and attention. I will, at
the appropriate time, be happy to answer questions.
[The prepared statement of Dr. Reed follows:]
Prepared Statement of Daniel A. Reed
Good morning, Mr. Chairman and Members of the Committee. Thank you
very much for granting me this opportunity to comment on appropriate
paths for scientific computing. I am Daniel Reed, Director of the
National Center for Supercomputing Applications (NCSA), one of three
NSF-funded high-end computing centers. I am also a researcher in high-
performance computing and a former head of the Department of Computer
Science at the University of Illinois.
In response to your questions, I would like to make three points
today regarding the status and future of advanced scientific computing.
1. Success and Scientific Opportunities
First, the National Science Foundation has played a critical role
in providing high-end computational infrastructure for the broad
university-based science and engineering community, a critical role
that must continue. NCSA began with NSF funding for the supercomputer
centers program in the 1980s. Via this program and its successors, the
NSF supercomputing centers have provided community access to high-end
computers that previously had been available only to researchers at
national laboratories. These supercomputing investments have not only
enabled scientific discovery, they have also catalyzed development of
new technologies and economic growth.
The Internet sprang from early DARPA investments and from funding
for NSFNet, which first connected NSF's supercomputing centers and
provided open access to high-end computing facilities. NCSA Mosaic, the
first graphical web browser, which spawned the web revolution, grew
from development of tools to support collaboration among distributed
scientific groups. Research by NCSA and the other supercomputing
centers was instrumental in the birth of scientific visualization--the
use of graphical imagery to provide insight into complex scientific
phenomena. Via ONR-funded security center, NCSA is creating new
cybersecurity technologies to safeguard the information infrastructure
of our nation and our military forces.
Today, the NSF supercomputing centers and their Partnerships for
Advanced Computational Infrastructure (PACI) program and Extended
Terascale Facility (ETF) partners are developing new tools for
analyzing and processing the prodigious volumes of experimental data
being produced by new scientific instruments. They are also developing
new Grid technologies that couple distributed instruments, data
archives and high-end computing facilities, allowing national research
collaborations on an unprecedented scale.
Access to high-end computing facilities at NCSA, the San Diego
Supercomputing Center (SDSC) and the Pittsburgh Supercomputing Center
(PSC) is provided by national peer review, with computational science
researchers reviewing the proposals for supercomputing time and
support. This process awards access to researchers without regard to
the source of their research funds. The three centers support
researchers with DOE, NIH and NASA awards, among others. This ensures
that the most promising proposals from across the entire science and
engineering community gain access to high-end computing hardware,
support staff and software. Examples include:
Simulation of cosmological evolution to test basic
theories of the large-scale structure of the universe
Quantum chromodynamics simulations to test the
Standard Model of particle physics
Numerical simulation of weather and severe storms for
accurate prediction of severe storm tracks
Climate modeling to understand the effects of climate
change and assess global warming
Studying the dynamics and energetics of complex
macromolecules for drug design
Nanomaterials design and assessment
Fluid flow studies to design more fuel-efficient
aircraft engines
From these, let me highlight just two examples of scientific
exploration enabled by high-end computing, and the limitations on
computing power we currently face.
A revolution is underway in both astronomy and high-energy physics,
powered in no small part by high-end scientific computing. New data,
taken from high-resolution sky surveys, have exposed profound questions
about the nature and structure of the universe. We now know that the
overwhelming majority of the matter in the universe is of an unknown
type, dubbed ``dark matter,'' that is not predicted by the Standard
Model of physics. Observations also suggest that an unknown ``dark
energy'' is causing the universe to expand at an ever-increasing rate.
Both of these discoveries are profound and unexpected.
Large-scale simulations at NCSA and other high-end computing
centers are being used to investigate models of the evolution of the
universe, providing insight in the viability of competing theories. The
goal of these simulations is to create a computational ``universe in a
box'' that evolves from first principles to conditions similar to that
seen today. These cosmological studies also highlight one of the unique
capabilities of large-scale scientific simulation--the ability to model
phenomena where experiments are otherwise not possible.
As a second example, the intersecting frontiers of biology and
high-end computing are illuminating biological processes and medical
treatments for disease. We have had many successes and others are near.
For example, given large numbers of small DNA fragments, high-end
computing enabled ``shotgun sequencing'' approaches for assembling the
first maps of the human genome. High-end computing has also enabled us
to understand how water moves through cell walls, how blood flow in the
heart can disrupt the plaque that causes embolisms, and how new drugs
behave.
Unraveling the DNA code for humans and organisms has enabled
biologists and biomedical researchers to ask new questions, such as how
genes control protein formation and cell regulatory pathways and how
different genes increase the susceptibility to disease. Biophysics and
molecular dynamics simulations are now being used to study the
structure and behavior of biological membranes, drug receptors and
proteins. In particular, understanding how proteins form three-
dimensional structures is central to designing better drugs and
combating deadly diseases such as HIV and SARS. Today's most powerful
high-end computing systems can only simulate microseconds of the
protein folding process; complete folding takes milliseconds or more.
Such large-scale biological simulations will require vast increases in
computing capability, perhaps as much as 1000 times today's capability.
Simply put, we are on the threshold of a new era of scientific
discovery, enabled by computational models of complex phenomena. From
astronomy to zoology, high-end computing has become a peer to theory
and experiment for exploring the frontiers of science and engineering.
This brings me to my second point: the challenges before us.
2. Challenges
Although large-scale computational simulation has assumed a role
equal to experiment and theory in the scientific community, within that
community, we face critical challenges. There is a large and unmet
demand for access to high-end computing in support of basic scientific
and engineering research. There are neither enough high-end computing
systems available nor are their capabilities adequate to address fully
the research challenges and opportunities. This view is supported by
recent workshops, reports and surveys, including the NSF report on
high-end computing and cyberinfrastructure, the DOD integrated high-end
computing report, and the DOE study on a science case for large-scale
simulation.
On behalf of the interagency High-End Computing Revitalization Task
Force (HECRTF), I recently chaired a workshop to gain community input
on high-end computing opportunities and needs. At the workshop,
researchers from multiple disciplines made compelling cases for
sustained computing performance of 50-100X beyond that currently
available. Moreover, researchers in every discipline at the HECRTF
workshop cited the difficulty in achieving high, sustained performance
(relative to peak) on complex applications to reach new, important
scientific thresholds. Let me cite just a few examples to illustrate
both the need and our current shortfall.
In high-energy physics, lattice quantum chromodynamics (QCD)
calculations, which compute the masses of fundamental particles from
first principles, require a sustained performance of 20-50 teraflops/
second (one teraflop is 1012 arithmetic operations per
second). This would enable predicative calculations for ongoing and
planned experiments. In magnetic fusion research, sustained performance
of 20 teraflops/second would allow full-scale tokamak simulations,
providing insights into the design and behavior of proposed fusion
reactor experiments such as the international ITER project. HECRTF
workshop participants also estimated that a sustained performance of 50
teraflops/second would be needed to develop realistic models of complex
mineral surfaces for environmental remediation, and to develop new
catalysts that are more energy efficient.
Note that each of these requirements is for sustained performance,
rather than peak hardware performance. This is notable for two reasons.
First, the aggregate peak performance of the high-end computing systems
now available at NSF's three supercomputing centers (NCSA, SDSC and
PSC) and DOE's National Energy Research Scientific Computing Center
(NERSC) is roughly 25 teraflops.\1\ Second, researchers in every
discipline at the HECRTF workshop cited the difficulty in achieving
high, sustained performance (relative to peak) on complex applications.
---------------------------------------------------------------------------
\1\ According to the most recent ``Top 500'' list (www.top500.org),
6 teraflops at PSC, 6 teraflops at NCSA, 3.7 teraflops at SDSC and 10
teraflops at NERSC. Upcoming deployments at NCSA, SDSC and PSC will
raise this number, but the aggregate will still be far less than user
requirements.
---------------------------------------------------------------------------
Simply put, the Nation's aggregate, open capability in high-end
computing is at best equal to the scientific community's estimate of
that needed for a single, breakthrough scientific application study.
This is an optimistic estimate, as it assumes one could couple all
these systems and achieve 100 percent efficiency. Instead, these open
systems are shared by a large number of users, and the achieved
application performance is often a small fraction of the peak hardware
performance. This is not an agency-specific issue, but rather a
shortfall in high-end computing capability that must be addressed by
all agencies to serve their community's needs.
Achieving high-performance for complex applications requires a
judicious match of computer architecture, system software and software
development tools. Most researchers in high-end computing believe the
key reasons for our current difficulties in achieving high performance
on complex scientific applications can be traced to (a) inadequate
research investment in software and (b) use of processor and memory
architectures that are not well matched to scientific applications.
Today, scientific applications are developed with software tools that
are crude compared to those used in the commercial sector. Low-level
programming, based on message-passing libraries, means that application
developers must provide deep knowledge of application software behavior
and its interaction with the underlying computing hardware. This is a
tremendous intellectual burden that, unless rectified, will continue to
limit the usability of high-end computing systems, restricting
effective access to a small cadre of researchers.
Developing effective software (programming languages and tools,
compilers, debuggers and performance tools) requires time and
experience. Roughly twenty years elapsed from the time vector systems
such as the Cray-1 first appeared in the 1970s until researchers and
vendors developed compilers that could automatically generate software
that operated as efficiently as that written by a human. This required
multiple iterations of research, testing, product deployment and
feedback before success was achieved.
In the 1990s, the U.S. HPCC program supported the development of
several new computer systems. In retrospect, we did not learn the
critical lesson of vector computing, namely the need for long-term,
sustained and balanced investment in both hardware and software. We
under-invested in software and expected innovative research approaches
to high-level programming to yield robust, mature software in only 2-3
years. One need only look at the development history of Microsoft
WindowsTM to recognize the importance of an iterated cycle of
development, deployment and feedback to develop an effective, widely
used product. High quality research software is not cheap, it is labor
intensive, and its successful creation requires the opportunity to
incorporate the lessons learned from previous versions.
The second challenge for high-end computing is dependence on
products derived too narrowly from the commercial computing market.
Although this provides enormous financial leverage and rapid increases
in peak processor performance, commercial and scientific computing
workloads differ in one important and critical way--access to memory.
Most commercial computer systems are designed to support applications
that access a small fraction of a system's total memory during a given
interval.
For commercial workloads, caches--small, high-speed memories
attached to the processor--can hold the critical data for rapid access.
In contrast, many, though not all, scientific applications (and several
of those critical to signals intelligence and cryptanalysis) have
irregular patterns of access to a large fraction of a system's memory.
This is not a criticism of vendors, but rather a marketplace reality we
must recognize and leverage. New high-end computing designs are needed
to support these characteristics, both for fully custom high-end
computer designs and more appropriate designs based on commodity
components.
The dramatic growth of the U.S. computing industry, with its
concomitant economic benefits, has shifted the balance of influence on
computing system design away from the government to the private sector.
As the relative size of the high-end computing market has shrunk, we
have not sustained the requisite levels of innovation and investment in
high-end architecture and software needed for long-term U.S.
competitiveness. Alternative strategies will be required.
This leads me to my third and final point: appropriate models of
cooperation and support for high-end computing.
3. Actions
We must change the model for development, acquisition and
deployment of high-end computing systems if the U.S. is to sustain the
leadership needed for scientific discovery and national security in the
long-term. The Japanese Earth System Simulator is a wakeup call, as it
highlights the critical importance of both industry-government
collaboration and long-term sustained investment. Reflecting the
lessons of long-term investment I discussed earlier, the Earth System
Simulator builds on twenty years of continued investment in a
particular hardware and software model, and the lessons of six product
generations. To sustain U.S. leadership in computational science, we
must pursue two concurrent and mutually supporting paths, one short- to
medium-term and the second long-term.
In the short- to medium-term, we must acquire and continue to
deploy additional high-end systems at larger scale if we are to satisfy
the unmet demand of the science and engineering research community.
NSF's recent cyberinfrastructure report,\2\ DOD's integrated high-end
computing report, and DOE's ultrascale simulation studies have all made
such recommendations. As one example, the cyberinfrastructure report
noted ``The United States academic research community should have
access to the most powerful computers that can be built and operated in
production mode at any point in time, rather than an order of magnitude
less powerful, as has often been the case in the past decade.'' The
cyberinfrastructure report estimated this deployment as costing roughly
$75M/year per facility, with $50M/year per facility allocated to high-
end computing hardware.
---------------------------------------------------------------------------
\2\ ``Revolutionizing Science and Engineering Through
Cyberinfrastructure: Report of the National Science Foundation Blue-
Ribbon Advisory Panel on Cyberinfrastructure,'' January 2003,
www.cise.nsf.gov/evnt/reports/toc.htm
---------------------------------------------------------------------------
Given the interdependence between application characteristics and
hardware architecture, this will require deployment of high-end systems
based on diverse architectures, including large-scale message-based
clusters, shared memory systems (SMPs) and vector systems.\3\ Moreover,
these systems must not be deployed in a vacuum, but rather must
leverage another critical element of sustainable infrastructure--the
experienced support staff members who work with application scientists
to use high-end systems effectively. These high-end centers must also
interoperate with a broad infrastructure of data archives, high-speed
networks and scientific instruments.
---------------------------------------------------------------------------
\3\ This is the approach we have adopted at NCSA, deploying
multiple platforms, each targeting a distinct set of application needs.
---------------------------------------------------------------------------
High-end computing system deployments should not be viewed as an
interagency competition, but rather as an unmet need that requires
aggressive responses from multiple agencies. NSF and its academic
supercomputing centers have successfully served the open academic
research community for seventeen years; NSF should build on this
success by deploying larger systems for open community access.
Similarly, DOE has well served the high-end computing needs of
laboratory researchers; it too should build on its successes. NIH, DOD,
NASA, NSA and other agencies also require high-end capabilities in
support of their missions, both for research and for national needs.
The need is so large, and the shortfall is so great, that broader
investment is needed by all agencies.
Concurrent with these deployments, we must begin a coordinated
research and development effort to create high-end systems that are
better matched to the characteristics of scientific applications. To be
successful, these efforts must be coordinated across agencies in a much
deeper and tighter way than in the past. This will require a broad,
interagency program of basic research into computer architectures,
system software, programming models, software tools and algorithms.
In addition, we must fund the design and construction of large-
scale prototypes of next-generation high-end systems that includes
balanced exploration of new hardware and software models, driven by
scientific application requirements. Multiple, concurrent efforts will
be required to reduce risk and to explore a sufficiently broad range of
ideas; six efforts, each federally funded at a minimum level of $5M-
$10M/year for five years, is the appropriate scale. At smaller scale,
one will not be able to gain the requisite insights into the interplay
of application needs, hardware capabilities, system software and
programming models.
Such large-scale prototyping efforts will require the deep
involvement and coordinated collaboration of vendors, national
laboratories and centers, and academic researchers, with coordinated,
multi-agency investment. After experimental assessment and community
feedback, the most promising efforts should then transition to even
larger scaling testing and vendor productization, and new prototyping
efforts should be launched. It is also important to remember the lesson
of the Earth System Simulator--the critical cycle of prototyping,
assessment, and commercialization must be a long-term, sustaining
investment, not a one time, crash program.
I believe we face both great opportunities and great challenges in
high-end computing. Scientific discovery via computational science
truly is the ``endless frontier'' of which Vannevar Bush spoke so
eloquently in 1945. The challenges are for us to sustain the research,
development and deployment of the high-end computing infrastructure
needed to enable those discoveries.
In conclusion, Mr. Chairman, let me thank you for this committee's
longstanding support for scientific discovery and innovation. Thank you
very much for your time and attention. I would be pleased to answer any
questions you might have.
Biography for Daniel A. Reed
Edward William and Jane Marr Gutgsell Professor; Director, National
Center for Supercomputing Applications; Director, National
Computational Science Alliance; Chief Architect, NSF ETF TeraGrid;
University of Illinois at Urbana-Champaign
Dan Reed serves as Director of the National Computational Science
Alliance (Alliance) and the National Center for Supercomputing
Applications (NCSA) at the University of Illinois, Urbana-Champaign. In
this dual directorship role, Reed provides strategic direction and
leadership to the Alliance and NCSA and is the principal investigator
for the Alliance cooperative agreement with the National Science
Foundation.
Dr. Reed is one of two principal investigators and the Chief
Architect for the NSF TeraGrid project to create a U.S. national
infrastructure for Grid computing. The TeraGrid is a multi-year effort
to build and deploy the world's largest, fastest, distributed computing
infrastructure for open scientific research. Scientists will use the
TeraGrid to make fundamental discoveries in fields as varied as
biomedicine, global climate, and astrophysics. Dr. Reed is also the
principal investigator and leader of NEESgrid, the system integration
project for NSF's George E. Brown Jr. Network for Earthquake
Engineering Simulation (NEES), which is integrating distributed
instruments, computing systems, and collaboration infrastructure to
transform earthquake engineering research.
Reed was head of the University of Illinois computer science
department from 1996 to 2001 and, before becoming NCSA and Alliance
director was co-leader of the Alliance's Enabling Technologies teams
for three years. He is a member of several national collaborations,
including the NSF Center for Grid Application Development Software, the
Department of Energy (DOE) Scientific Discovery through Advanced
Computing program, and the Los Alamos Computer Science Institute. He is
chair of the NERSC Policy Board for Lawrence Berkeley National
Laboratory, is co-chair of the Grid Physics Network Advisory Committee
and is a member of the board of directors of the Computing Research
Association (CRA). He recently served as program chair for the workshop
on the Road Map for the Revitalization of High End Computing for
Information Technology Research and Development (NTTRD). He is an
incoming member of the President's Information Technology Advisory
Committee (PITAC). In addition, he served as a member of Illinois
Governor's VentureTECH committee, which advised the former governor on
technology investment in Illinois.
Chairman Boehlert. Thank you, Dr. Reed. And I might say you
lived up to the advanced billing so ably presented by Mr.
Johnson.
Dr. Reed. Thank you.
Chairman Boehlert. Mr. Scarafino.
STATEMENT OF MR. VINCENT F. SCARAFINO, MANAGER, NUMERICALLY
INTENSIVE COMPUTING, FORD MOTOR COMPANY
Mr. Scarafino. Thank you for the opportunity to testify
regarding the national needs for advanced scientific computing
and industrial applications. My name is Vincent Scarafino, and
I am Manager of Numerically Intensive Computing for Ford Motor
Company.
Ford has a long and proud history of leadership in
advancing engineering applications and technologies that
stretches back over the last 100 years. Ford uses high-
performance computing to help predict the behavior of its
products in nominal and extreme conditions. Approximately half
of the computing capacity is used for safety analysis to
determine what its vehicles will do in impact situations. Other
uses include vehicle ride performance, usually referred to as
NVH, noise, vibration, and harshness, which is predicted using
model analysis techniques. Fluid flow analysis is used to
predict such things as drag characteristics of vehicles, under
hood temperatures, defroster and interior climate control,
catalytic converter design, and exhaust systems.
Computing capabilities allow Ford to accelerate the design
cycle. Better high-end computing systems will help engineers
balance competing design requirements. Performance, durability,
crash worthiness, occupant and pedestrian protection are among
them. These tools are necessary to stay in business.
Competition from national and international companies is
intense.
Significantly faster high-end machines would help improve
the ability to predict vehicle safety as well as the durability
and wind noise characteristics, which are among the most cited
customer issues. Although safety codes now run well on--in
parallel on commodity-based clusters, high-end machines would
provide improved turnaround times, leading to reduced design
cycle time and the opportunity to explore a greater variance of
design parameters. Advanced durability and wind noise analyses
typically require computer runs greater than two weeks, and
this is with coarse models on the fastest machines we have
access to. The durability work runs on 4-way SMP 1.25 GHz Alpha
processors. The wind noise runs with 6-way SMP on a Cray T90.
They cannot effectively use more processors.
Up until the mid-1990's, the Federal Government had helped
with the development of high-end machines with faster, more
powerful processing capability and matching memory bandwidth
and latency characteristics by helping to fund development and
create a market for them. These machines were built mainly to
meet the needs of government security and scientific research.
Once they were built, there was a limited, but significant
application of these machines in the private sector. The
availability of higher capability machines advanced the
application of science in the private sector. This worked quite
well.
In the mid-1990's, the Federal Government decided to rely
on utilizing off-the-shelf components and depend on the ability
to combine thousands of these components to work in harmony to
meet its advanced high-performance computing needs. The result
was an advance in the areas of computer science that dealt with
parallel processing. Over the last eight years, some kinds of
applications have adapted well to the more constrained
environment supported by these commodity-based machines.
Vehicle safety analysis programs are an example. Most vehicle
impact analysis can now be done on commodity clusters. We have
learned how to do ``old science'' considerably cheaper.
We have not made any significant advancement in new
science, some examples of which would include advanced occupant
injury analysis and the modeling of new complex materials, such
as composites. The physics is more complex, and the
computational requirements are beyond current capability. The
hardest problems do not adapt well to parallel architectures.
Either we don't know enough about the problem to develop a
parallel solution, or they do not--they are not parallel by
nature.
The Federal Government cannot rely on fundamental economic
forces to advance high-performance computing capability. The
only economic model with enough volume to drive this kind of
development is the video game industry. Unfortunately, these
models--these machines are very fast only for a particular
application and do not provide a general solution. The Federal
Government should help with the advancement of high-end
processor design and other fundamental components necessary to
develop well-balanced, highly capable machines. U.S. leadership
is currently at risk.
The advanced computing needs of the scientific community
and the private sector are not diverging. The fact that there
has been no fundamental advance in high-performance capability
computers in the last eight years has forced these communities
to adapt less qualified commercial offerings to the solution of
their problems. If advanced computing capability becomes
available in the form of next-generation supercomputers, the
scientific community and the private sector would be able to
utilize them for the application of new science.
Thank you.
[The prepared statement of Mr. Scarafino follows:]
Prepared Statement of Vincent F. Scarafino
Thank you for the opportunity to testify regarding the national
needs for advanced scientific computing and industrial applications. My
name is Vincent Scarafino and I am Manager of Numerically Intensive
Computing for Ford Motor Company. Our automotive brands include Ford,
Lincoln, Mercury, Volvo, Jaguar, Land Rover, Aston Martin and Mazda.
Ford has a long and proud history of leadership in advanced
engineering applications and technologies that stretches back over the
last 100 years. Henry Ford set the tone for the research and
development that has brought many firsts to the mass public. In the
future, the company will continue to apply innovative technology to its
core business. New technologies, such as hybrid electric and fuel cell
powered vehicles, will help achieve new goals, reward our shareholders
and benefit society. We understand that the quality and safety or our
products are fundamental to our corporate success.
For example, to be among the leaders in vehicles safety, we must
commit ourselves to ongoing improvement of the safety and value our
products. We do this through research, product development and
extensive testing of our products.
Ford uses high-performance computing to help predict the behavior
of its products in nominal and extreme conditions. Approximately half
of the computing capacity is used for safety analysis to determine what
its vehicles will do in impact situations. Other uses include vehicle
ride performance (usually referred to as NVH--noise, vibration and
harshness), which is predicted using modal analysis techniques. Fluid
flow analysis is used to predict such things as drag characteristics of
vehicles, under hood temperatures, defroster and interior climate
control, catalytic converter design, and exhaust systems.
Computing capabilities allow Ford to accelerate the design cycle.
Better high-end computing systems will help engineers balance competing
design requirements--performance, durability, crash-worthiness,
occupant and pedestrian protection are among them. These tools are
necessary to stay in business. Competition from national and
international companies is intense.
Significantly faster high-end machines would help improve the
ability to predict vehicle safety as well as durability and wind noise
characteristics, which are among the most cited customer issues.
Although safety codes now run well in parallel on commodity based
clusters, high-end machines would provide improved turnaround times
leading to reduced design cycle time and the opportunity to explore a
greater variance of design parameters. Advanced durability and wind
noise analyses typically require computer runs greater than two weeks,
and this is with coarse models on the fastest machines we have access
to. The durability work runs on 4-way SMP 1.25 GHz Alpha processors.
The wind noise runs with 6-way SMP on a Cray T90. They cannot
effectively use more processors.
Up until the mid 1990's, the Federal Government had helped with the
development of high-end machines with faster, more powerful processing
capability and matching memory bandwidth and latency characteristics by
helping to fund development and create a market for them. These
machines were built mainly to meet the needs of government security and
scientific research. Once they were built, there was a limited, but
significant application of these machines in the private sector. The
availability of higher capability machines advanced the application of
science in the private sector. This worked quite well.
In the mid 1990's the Federal Government decided to rely on
utilizing off-the-shelf components and depend on the ability to combine
thousands of these components to work in harmony to meet its advanced
high-performance computing needs. The result was an advance in the
areas of computer science that dealt with parallel processing. Over the
last eight years, some kinds of applications have adapted well to the
more constrained environment supported by these commodity based
machines. Vehicle safety analysis programs are an example. Most vehicle
impact analysis can now be done on commodity clusters. We have learned
how to do ``old science'' considerably cheaper. We have not made any
significant advancement in new science. Some examples include advanced
occupant injury analysis and the modeling of new complex materials such
as composites. The physics is more complex and the computational
requirements are beyond current capability. The hardest problems do not
adapt well to parallel architectures. Either we don't know enough about
the problem to develop a parallel solution, or they are not parallel by
nature.
The Federal Government cannot rely on fundamental economic forces
to advance high-performance computing capability. The only economic
model with enough volume to drive this kind of development is the video
game industry. Unfortunately these machines are very fast only for a
particular application and do not provide a general solution. The
Federal Government should help with the advancement of high-end
processor design and other fundamental components necessary to develop
well-balanced, highly capable machines. U.S. leadership is currently at
risk.
The advanced computing needs of the scientific community and the
private sector are not diverging. The fact that there has been no
fundamental advance in high-performance capability computers in the
last eight years has forced these communities to adapt less qualified
commercial offerings to the solution of their problems. If advanced
computing capability becomes available in the form of next generation
supercomputers, the scientific community and the private sector would
be able to utilize them for the application of new science.
Ford continues to invest time, money and a significant portion of
our company's human resources to study and explore how our vehicles can
perform even better for the next 100 years. We believe that the next
generation of supercomputers is essential to ensure that our industry,
which has contributed greatly to economic growth in this country,
remains competitive.
Biography for Vincent F. Scarafino
Vince Scarafino is currently Manager of Numerically Intensive
Computing at Ford Motor Company. He has been with Ford for close to 30
years and has over 35 years of experience with computers and computing
systems. Most of his work has been in the engineering and scientific
areas. He started with GE 635, IBM 7090, and IBM 1130 machines. His
experience with security and relational data base technology on Multics
systems lead to sharing knowledge with academic and government
organizations. His efforts in the last decade have been dedicated to
providing flexible and reliable supercomputer resources for the vehicle
product development community at Ford via an open systems-based grid
environment connecting three data centers on two continents. He
currently manages a broad range of high performance computers, from
Cray T90 to Linux Beowulf Clusters, that are used by Ford's engineers
around the world for such things as vehicle design and safety analysis.
Vince has a degree in mechanical engineering from the University of
Michigan.
Discussion
Chairman Boehlert. Thank you very much.
Let me start by asking Dr. Orbach and Dr. Freeman. This
interagency coordination that we think is so vital, as I
understand it, that is sort of being shepherded by OSTP. And
how long has that been in operation, this interagency
coordinating vehicle? What are they called? Council or--Dr.
Orbach?
Dr. Orbach. It is the High-End Computing Revitalization
Task Force. That Committee is expected to finish its report in
August, next month, and has been in progress for about six
months.
Dr. Freeman. A bit less, I believe. It started in the April
time frame, I believe.
Chairman Boehlert. Okay. But the report is not the end of
the job. I mean, they are going to--it is going to continue,
isn't it?
Dr. Freeman. I am not certain of the intentions of OSTP.
Chairman Boehlert. Do you think it should continue? In
other words, don't you think we need an interagency
coordinating vehicle to carry forward on a continuing basis a
program where the principals of the agencies involved get
together, not to just chat and pass the time of day, but get
really serious about a really serious subject?
Dr. Freeman. I certainly agree that coordination is
necessary. Let me remind you that there has been in existence,
for over 10 years, the National Coordination Office for
Networking and IT R&D. Indeed, that office, the National
Coordination Office, is managing the High-End Computing
Revitalization Task Force that we were just discussing.
Chairman Boehlert. That has been in existence for a decade
at a time when--instead of going like this, as we all want to,
and the recorder can't say ``like this'', so let me point out a
dramatic upward turn, it sort of leveled off. And we are
getting the pants beat off of us in some areas.
Dr. Freeman. I think my point, Mr. Chairman, is that the
mechanism is in place for coordination. As you know, I have
only been in the government for about a year, so I can't speak
to the total history of that office, but I think, certainly--
and I chair or co-chair the Interagency Working Group. But
the--what I hear you calling for is a higher level
coordination. I think some steps have already been taken in
that direction that Dr. Orbach may be able to speak to.
Chairman Boehlert. But would you agree that OSTP is the
place, maybe, to be the coordinating vehicle for this
interagency cooperative effort? And don't be turf-conscious. I
mean, just sort of give it some thought. I mean, the Science
Advisor to the President, head of the Office of Science and
Technology Policy, it would seem to me that would be the ideal
vehicle and that the principals not send a designee to that
meeting, or at least every meeting. Once in a while the
principals, including Dr. Marberger and Dr. Freeman and Dr.
Orbach, you know, you get there, not some designee. And you
really talk turkey.
And then it seems to me that as you go in your budget
preparation for the next years, you are going into OMB. You go
so--forward with a coordinated plan, so that if OMB is looking
at Orbach's recommendation for DOE, they will be mindful of
what Freeman's recommendation is for NSF and what somebody
else's is for DARPA, in other words, a coordinated approach.
This is too big a challenge to not just really put our shoulder
to the wheel, so to speak, and that is sort of an uncomfortable
position, but you know exactly what I am saying.
Dr. Orbach, do you have any comment on that?
Dr. Orbach. Mr. Chairman, I subscribe to your remarks
directly. In fact, the principals did meet, under Dr.
Marberger's direction, a few weeks ago to discuss precisely the
issues, which you laid out.
Chairman Boehlert. And they are going forward?
Dr. Orbach. And they are going forward. And my personal
preference would be that OSTP would continue that relationship
and leadership. When the principals met, there was unanimity
that it was--needed to be a collaborative affair where each of
us brought our areas of expertise to the table and work
together.
Chairman Boehlert. And the coordinated approach to the
budget process is extremely important. Speaking about the
budget process, what about beyond '04, Dr. Freeman, with NSF's
plans for the Supercomputing Centers? Where do you envision
going in '05 and '06?
Dr. Freeman. Well, as I noted in my both verbal and written
testimony, although we have announced plans to change the
modality of funding in line with the recommendations of the
Atkins Report, we intend that the amount of money going into
the activities that are currently covered by that PACI program,
in fact, will increase. We have announced some aspects of that.
As I noted, other aspects are currently under internal
preparation and will be announced and vetted through our
National Science Board in the coming month.
Chairman Boehlert. So is that in the future I should be
optimistic about our centers beyond '04?
Dr. Freeman. Absolutely, and I----
Chairman Boehlert. And I should be optimistic that
resources will come to do some of the things that need to be
done in terms of purchasing hardware, for example?
Dr. Freeman. That is certainly our intention.
Chairman Boehlert. Thank you very much. And Mr. Johnson, if
he were here, would be comforted by that answer.
Let me ask----
Dr. Freeman. As well as Dr. Reed, I might add.
Chairman Boehlert. Well, I know. I didn't want to be too
obvious.
Let me turn to our non-government witnesses, Dr. Reed and
Mr. Scarafino. And I thank you for your input. Is the Federal
Government policy, as you understand it, and I am not quite
sure it is crystal clear, do you think we are moving in the
right direction? And if not, where would you put some
additional focus? We will start with you, Dr. Reed.
Dr. Reed. Well, I agree, absolutely, that I think greater
coordination is needed. I think one of the challenges for us--
as I said, there are really two aspects of this problem. There
is what we do in the short-term and the medium-term to address
the shortfall of the scientific needs, vis-a-vis the Earth
Simulator and what is available to the science community. And
there is a clear shortfall. And I realize everyone that comes
before you pleads poverty and asks for more money. I mean, that
is--I know that. But there is a clear need for additional
resources to address the community. I think it is fair to say
that in terms of one sort of breakthrough calculation, we,
perhaps, have, in the open community, enough supercomputing
capability for, perhaps, solving one problem, at the moment.
And there are many problems at that scale that the community
would like to solve. So that is one of the issues.
The longer-term issue, and the place where I think
coordination is equally critical in addressing the issue that
Mr. Scarafino addressed as well, and that is the commercial
application of technology to high-performance computing. There
is a disconnect in the capability that we can procure,
regardless of the amount of money we have, to address certain
needs. And some of those machines simply aren't available at
any price right now.
And the place where coordination would be enormously
helpful is not only--if you think about the research pipeline
from the basic long-term research that is required to
investigate the new architectures and new software and new
algorithms for applications, the place where we have had a gap,
in some sense, has been the building of machines at
sufficiently large scale to test them so that industry would
pick them up and commercialize them and make them products.
That gap is a place where coordination across DARPA, DOE, NSF,
and other agencies could really play a critical role. And I
think one of the things we have to do there is fund coordinated
teams across those agencies to address some of those
challenges. That is a place where, operationally, a very big
difference could be made. And that was a message that was
echoed very loudly at the workshop I chaired, which was the
community input workshop to the process that Dr. Freeman and
Dr. Orbach discussed where the science and engineering
community was providing responses to what the tentative agency
plans were.
Chairman Boehlert. Mr. Scarafino?
Mr. Scarafino. I am not sure exactly what kind of
mechanisms could be best used to accomplish these things, but
it is very good that the Federal Government is awakened to the
shortcomings in the very-high computing arena and that they are
basically taking steps to correct that. That is----
Chairman Boehlert. So I would imagine within Ford--you
notice I gave you a plug in my introductory statement. I said
when the light bulb went on, we had a better idea. Thank you
very much for that response.
The Chair recognizes Mr. Bell.
Mr. Bell. Thank you, Mr. Chairman. And thank you for
calling this hearing. I would agree with the Chair and Ranking
Member that it is an extremely important topic, and I
appreciate your testimony here today.
I want to follow up on an issue that Chairman Boehlert
raised in his opening statement regarding Japan and the fact
that they are recognized as the world leader in supercomputing
and certainly have a lead on us and combine that with another
subject that this committee has paid a great deal of attention
to over the course of this year, and that is nanotechnology,
because in a similar parallel, Japan is also a world leader in
the area of nanotech. As you all know, in the future, it is
likely that nanotech particles or a nanochip will be used to
replace the silicon chip, and nanotech products or nanochips
are smaller and can hold more memory and have a lot of
advantages. And since it is the role of the government to
continue to fund research on new approaches to building high-
performance computing hardware or supercomputers and nanotech
seems to be a means to this end, can you tell me, and perhaps
beginning with Dr. Orbach and anybody else who would like to
comment, in what ways the government is looking at
nanotechnology to create or develop new supercomputing hardware
and any progress that is being made in that area?
Dr. Orbach. It works both ways. Not only can nanotechnology
assist in the creation of new architectures, but also the high-
end computing is required to understand the properties of
nanoscale materials. The Office of Science had a workshop on
theory and modeling in nanoscience, and discovered that
computational capabilities, exactly as Mr. Scarafino talked
about with regard to composites, is essential to really
understand what it is you are doing when you are creating these
new materials at the atomic level. So it works both ways and is
a critical need.
Mr. Bell. Dr. Reed, I saw you--or Dr. Freeman, did you want
to comment?
Dr. Freeman. Yes, I would just note that NSF is heavily
involved in the government-wide nanotechnology initiative, as
you know. One of the missions of the directorate that I head is
to explore, if I may use the vernacular, far out possibilities
for computation: quantum computing, biological computing in the
sense of using biological materials to compute. And I am sure
that some of our investigators are looking at some of the types
of applications that you referred to as well.
Mr. Bell. Dr. Reed?
Dr. Reed. I just want to echo what was said. It is
certainly the case that applications of high-end computing are
really critical to exploring novel materials. One of the things
that we are close to being able to do, that we can see on the
horizon, is looking at ab initio calculations from first
principles to design new composite materials. And that is one
of the applications.
To hop back to the lead-in of your question about the Earth
System Simulator and that aspect of it, I think one of the
lessons to draw from that effort is the importance of long-term
investment. The Earth Simulator was not an overnight sensation.
It was a long period of investment. There were roughly 20 years
of R&D behind that machine and six generations of precursor
products where ideas and technologies were combined to build
that. And that is why earlier in my oral testimony I said that
sustained investment is really part of the key to developing
these technologies.
But Dr. Freeman and Dr. Orbach are absolutely correct that
there is an interplay. It goes both ways. The application of
high-end computing can design materials, and new materials are
really critical in designing the next generation of machines.
Mr. Bell. Mr. Scarafino?
Mr. Scarafino. I don't really have any expertise in this
area, so there is nothing I could add.
Mr. Bell. Fair enough. And Dr. Reed, maybe I should ask
you, or any of you that know, what Japan is doing in regard to
nanotechnology as it relates to supercomputing and what the
U.S. needs to do, in your opinion, to catch up. Do you know?
Dr. Reed. I can't comment specifically on the
nanotechnology aspect. I can say that the Japanese have
historically had long-term investment in building a whole
series of machines of which the Earth Simulator is just one
example. One example of a machine that they have built at least
six generations of, as well, is focused on machines designed
specifically to support high-energy physics. There have been a
series of machines called GRAPE, up through, I believe, GRAPE-
6, that are focused on the same kinds of problems that we
looked at here in trying to compute from first principals the
masses of elementary particles. And those machines have been
designed working directly with the physicists who want to solve
those problems to understand the match of the application
requirements to the architecture and the software.
And that interplay of the scientific computing needs with
the design on the hardware and the software is where the--often
a disconnect has been when we have tried to leverage commercial
hardware, because it is driven, as Mr. Scarafino mentioned,
by--primarily by the needs of the commercial market. And that
interplay is one of the things that I think has been the
hallmark of successful scientific computing over the years at
the National Laboratories, at the centers. It has been looking
at the way hardware, architecture, software, driven by
applications, can shape the future.
Dr. Freeman. I would certainly agree with Dr. Reed. But I
would just add that I have a meeting this afternoon, indeed, to
help shape a study that we intend to kick off in cooperation
with several other agencies, specifically to make sure that we
are well advised as to what is going on in Japan in some of
these forefront areas.
Mr. Bell. And then we would have a better idea of what
would be required to----
Dr. Freeman. Precisely.
Mr. Bell. Dr. Orbach, any thoughts on that?
Dr. Orbach. I would second your observation. The Japanese
have been very aggressive in the nanotechnology area. And we
will provide, for the record, some of the details to bring you
up--all of us up to speed on that, if that would be acceptable.
[Note: This information is located in Appendix 2:
Additional Material for the Record.]
Mr. Bell. It would be, and I appreciate it.
And thank you, Mr. Chairman.
Chairman Boehlert. Thank you, Mr. Bell.
And now it is a pleasure to turn to a distinguished
scientist and educator, a fellow of the American Physical
Society, the gentleman from Michigan, Dr. Ehlers.
Mr. Ehlers. Thank you, Mr. Chairman. That is almost as
glowing as Dr. Reed's introduction.
I am fascinated listening to this and particularly the
discussions of teraflops, et cetera. I go back to the days when
I was happy to get one flop. I cut my eyeteeth on the IBM-650,
which I believe was developed before Dr. Reed was even born.
And I was delighted when--and I was one of the first to use
that in my field of physics and found it extremely useful.
I--two questions. The first one, I am pleased with the
memorandum of understanding that you referenced, Dr. Orbach, in
working with other agencies and the cooperation that is being
discussed between DOE and NNSA, both your branch and the NNSA,
as well as DARPA and NSF. But what about some of the other
agencies, for example, NOAA [National Oceanic and Atmospheric
Administration] and its affiliate NCAR [National Center for
Atmospheric Research], use--have a great need for very high-
performance computers. How are they involved in all that is
going on, and how do you share resources with them? Are they
part of this task force? Are they part of the decision-making
process? Because clearly, climate modeling is of an immense
interest, particularly in relationship to climate change, but
even in the mundane aspects of everyday weather projection. Dr.
Orbach, do you want to kick that off?
Dr. Orbach. Yes, we are working--the--very closely, the
Department of Energy Office of Science has a responsibility for
climate modeling among all of the other agencies. And we are
working very closely with NOAA and the other programs to create
computational architectures, which will help in modeling. It is
interesting to note that the Earth Simulator can do much better
predictions, long-term predictions, on climate than we can by
almost a factor of 10. Their grid size is that much smaller. So
it is just evidence of the computational power that these
machines can provide.
Mr. Ehlers. Dr. Freeman, anything to add?
Dr. Freeman. Let me just note, sir, that I believe you
mentioned NCAR. That is actually a federally-funded research
and development center that is largely supported by the
National Science Foundation. And so through that mechanism,
they are very intricately involved with our plans at NSF.
Likewise, in terms of the OSTP planning activity, I believe
NOAA is involved there, as are, essentially, all of the
agencies that you might expect to be engaged in this activity.
Mr. Ehlers. Thank you.
My second question, and that has to do with, Dr. Orbach,
with DOE's plans to provide supercomputer resources to academic
researchers. And I am wondering how this is going to relate to
the NSF Supercomputer Centers. Are--is there going to be a
meshing here, or is this simply going to provide another avenue
for academic researchers? They can knock on the door of a
Supercomputing Center. If they don't get what they need, they
knock on the door of DOE and say, ``What do you have to
offer?'' Could you amplify on how that is--how that
interrelationship is going to work out?
Dr. Orbach. Well, the interrelationship is very important,
because the National Science Foundation grid computation
structure will enable researchers everywhere in the United
States to couple into the high-end computational program. So we
rely on NSF for those relationships. We are going to be opening
our largest computer, NERSC, at Lawrence Berkeley National
Laboratory to grand challenge calculations. Ten percent of
NERSC will be available to any scientist, regardless of where
their support comes from, actually internationally, to compete
for a grand challenge calculation.
The opportunity to couple depends on the network of the
cyberinfrastructure, which you have heard referred to before,
and so we are working in a very interdependent way to enable
access to these machines.
Mr. Ehlers. Thank you.
My final question and concern is, first of all, what is
going on now is wonderful, but I think it is very late. A
number of years ago, the Japanese took the lead. We have
maintained our present more by a force of the Federal
Government saying, ``We are only going to buy American
computers rather than Japanese,'' even though we did get some
Japanese ones. But we lost our edge. How did that happen, and
is this really going to help us regain our edge in the
international arena? I will open it up to--Dr. Reed, you are
itching to answer this one.
Dr. Reed. I don't know if I am itching to answer, but I
will try.
I think one of the things that happened was that what
happened, if you look historically in--to hop back, perhaps not
to the 650, but certainly to earlier days, the machines that
were designed during the height of the Cold War were driven
clearly by national needs for security and weapons design. And
there was a compelling national need as well as a market
opportunity. As the computing market overall grew, high-
performance computing became a smaller fraction of the overall
market, and therefore the financial leverage that the Federal
Government had at the high-end has declined some. It is still
substantial and dominant, but it has declined. And that has led
manufacturers to use commercial products, or variations
thereof, to try to address those needs.
But I think the critical thing that happened was there is a
gap between basic research and next-generation architecture to
the single investigator model and production. And what happens
in that intermediate phase is testing ideas at sufficient scale
that vendors see that a productization of that idea is possible
and viable and likely to succeed. It is much less common for
ideas from papers to have a revolutionary effect on new-
generation designs. They have an evolutionary effect, and we
have seen that in commercial microprocessor design. But
something at sufficiently prototype scale that you can see with
real applications that there are benefits that you can get the
fractions of peak performance that machines, like the Earth
Simulator, have demonstrated.
That gap of testing those ideas at scale to provide the
push to get it over the energy barrier, if you will, to
productization is one of the things that we haven't had. And I
think if we could rekindle that effort, and it did very much go
on in the 1990's, even, with the HPCC [High Performance
Computing and Communications] Program, and before that, in the
'80's and '70's, we would go a long way to regaining the lead.
As I said, it is not a one-time crash effort. It is a
process that really needs to continue and be sustained at some
level to feed new ideas into the process.
Mr. Ehlers. And how did the Japanese get their edge?
Dr. Reed. They sustained an investment in machines. As we
looked at vector machines, the conventional wisdom, if you
will, was that that was not a path to continue. And in the
early 1990's, we decided, and I use ``we'' in the broad sense,
decided that there was not much more headroom in that
direction, and we started to explore other alternatives, which
led to the parallel computing market that Mr. Scarafino
referred to earlier. But we largely stopped development in
those vector machines that capitalized on very efficient access
to memory that supported scientific applications. Almost no
development happened there and little investment for a period
of 10 years or more.
The Japanese continued to invest in those machines to build
next-generation products. And that is why I said the Earth
Simulator is really the product of 20 years investment and at
least a half a dozen product generations. We went in a
different direction, which yielded some real benefits, but we
stopped going in a direction that still had additional
opportunity.
Mr. Ehlers. Thank you.
Chairman Boehlert. Thank you.
Mr. Udall.
Mr. Udall. Thank you, Mr. Chairman. I want to thank the
panel as well and apologize. I was at another hearing, and I
hope I am not redundant in my question. But I do want to pick
on my colleague, Mr. Ehlers' questions about parallel and
vector computing and direct the question to Dr. Freeman and Dr.
Orbach.
Are there fields in the area of science and engineering
where progress has been delayed because of inadequate access to
these high-end computing technologies? I am mindful of the
Clinton Administration when they did a national climate
assessment study as--in the last years of that Administration,
I believe--had to ask the indulgence of the Europeans and the
Canadians and the British for computing power to actually draw
that assessment on climate. That is an interesting dynamic, to
say the least. Would you all comment on my question and--if you
would?
Dr. Freeman. I think, certainly, some of the testimony this
morning and certainly it is the case that there are very
important problems that no one, including the Japanese, can
properly solve today. And there are certainly those that--where
our progress or our particular level of expertise may be
retarded because of our not having the, not only capability,
but the capacity, that is enough supercomputers to go around,
if you will.
Mr. Udall. Are there particular fields, Dr. Freeman or Dr.
Orbach that----
Dr. Orbach. Yes, I wanted to amplify on that. Indeed, there
are fields.
Mr. Udall. Um-hum.
Dr. Orbach. There are--we have had about eight virtual
workshops from across the scientific spectrum that have looked
at opportunities that high-end computation could provide. But I
would like to turn it around a little bit. I think the Japanese
are to be congratulated for construction of this machine. What
it did is it showed us that it was possible to get very high
sustained speeds on problems of scientific interest. They were
able to get 26.5 teraflops on geophysics problems. And as you
heard previously, we think that the range of somewhere between
25 to, maybe, 50 teraflops opens up a whole new set of
opportunities----
Mr. Udall. Um-hum.
Dr. Orbach [continuing]. For science that had never been
realized before. And so what we have, thanks to the Japanese
now, is an existence proof, namely that it is possible to
generate these high sustained speeds on specific scientific
questions. Now whether it will be one or two or three different
architectures, which will be required to address these, is what
we are investigating now. So the United States is pursuing
assiduously exploration of different architectures to see which
is best suited for the problems at hand.
Mr. Udall. Um-hum. Dr. Reed, you seem to be energized by
the question as well. Would you like to comment?
Dr. Reed. Well, there is a long list of application
domains. One of the things that came out at the workshop that I
mentioned earlier, as well as other workshops, has been talking
to application scientists about the opportunities that they
see. A couple of examples as illustration, one that is of
intellectual interest and one that is of very practical
interest. Intellectual interest, there have been huge new
results from observational astronomy that suggest that most of
the matter and energy that we see is, in fact, a small fraction
of the universe. And so trying to understand dark matter and
dark energy and build models of the large scale evolution of
the universe, in essence, to answer one of the really big
questions: how did everything get here?
Mr. Udall. Um-hum.
Dr. Reed. It is a problem that suddenly becomes possible.
And it is a compelling example of the power of high-performance
computing, because we can't do experiments on the universe. We
only have one. It is what it is. But we can evolve models of it
from first principles and compare those with observational
data. That is really one of the unique capabilities of
computational modeling, in comparison to experiment and theory.
Another one, of course, is looking forward to rational drug
design and being able to understand how proteins fold and how
various molecules can dock with those. Those are problems that
are out of range right now, but with additional computing
capability, have very practical benefits, because a huge
fraction of the expense that pharmaceutical companies incur is
in assessing potential drug candidates. They discard thousands
for every one that even reaches first stage clinical trials.
And high-performance computing can compress that time scale. It
can reduce the costs for them.
Mr. Udall. Who would own that computer that would be
utilized by the pharmaceutical companies?
Dr. Reed. Well, that is an interesting history there,
because it is a great example of the interplay between
government investment and industry. In fact, one of the very
first technology transfers that my center, NCSA, was engaged in
years ago, was where we transferred a lot of early vector
computing technology. They actually bought their own Cray
vector supercomputer and used for some early drug designs. So
there were connections and collaborations where that technology
can flow.
Mr. Udall. It sounds like a nice example of the public/
private partnership and public dollars going to public but also
private benefit. Thank you.
Chairman Boehlert. Thank you very much.
Mr. Udall. Thank you, Mr. Chairman.
Chairman Boehlert. The gentleman's time is expired.
It is a pleasure to welcome the newest Member of our
Committee, and a valued addition to the Committee, I might add,
Mr. Randy Neugebauer from Texas.
Mr. Neugebauer. Thank you, Mr. Chairman.
Mr. Reed, do you believe the NSF is providing adequate
support for the Supercomputer Centers? And have you detected
any change in policy with regard to NSF's commitment to the
Supercomputer Centers?
Dr. Reed. Well, as I said at the outset, NSF has been a
long and steadfast supporter of high-performance computing, yet
the centers have a 17-year history, as Chairman Boehlert
mentioned before. They were begun in the 1980's. And we look
forward to continued investment. We also recognize that NSF has
a portfolio of investments in infrastructure and research. And
budgets are constrained at the moment.
Having said that, we do certainly see untapped
opportunities where additional capability and investment, not
only in the hardware, but in the people and the support
infrastructure, could attack some new problems.
Mr. Neugebauer. Thank you.
Chairman Boehlert. Boy, was that a diplomatic answer.
Mr. Neugebauer. Exactly.
Dr. Reed, and Mr. Scarafino, Dr. Reed's testimony states
that there is a large and unmet demand for access to high-end
computing in support of basic scientific and engineering
research. Dr. Freeman and Dr. Orbach have described their
agencies' plans for the future. And this question could be to
Dr. Reed and to Mr. Scarafino. Do you believe that these plans
are adequate to provide more high-end computing systems and
then enable the development of the new capabilities?
Dr. Reed. Why--as I said before, and as you quoted from my
written testimony, I do think there is an unmet demand for
capability. At a high-level, I think it is fair to say that we
can perhaps solve one or two of these really critical problems.
As Mr. Bell asked about nanotechnology, a large-scale effort to
understand new nanomaterials itself could largely consume the
available open capacity that we have at the moment.
There are a whole set of problems like that, and it is a
matter of making priority choices about what the level of
investment is and where one will accelerate progress. There is
no doubt that additional investment would yield new progress,
however.
Mr. Scarafino. Since about the mid-1990's, we have kind of
been at a standstill with regard to the availability of more
capable computers. The area where we are struggling with as far
as making advances is in the area of durability and in the area
of wind noise. It has kind of been a--the concentration has
been on actually doing the stuff that we have been doing for
the last 10 years more efficiently. In about, I would say, 1996
or so, we had the expectation that the next-generation of the
Cray T90, which would have given us a boost of performance,
would have naturally been available, but ended up being
canceled.
So it is amazing that we still have a T90 at Ford. That
machine is almost 10 years old, and it still does things that,
basically, cannot be done effectively by anything new. So I am
looking forward to future advances in this--pushing this type
of technology so that we can address the difficult problems
that have kind of come to a standstill.
Mr. Neugebauer. As a follow-up to that, because part of the
question there is the access to the capability and then
developing the capability, is the problem at Ford the
capability or the access to existing capability?
Mr. Scarafino. The problem is that there is no place we can
go to buy something like that. So I--basically, it is really
not available, as far as--there isn't something out there that
we can't have access to. I mean, we can make arrangements to
get access to something, if it is available.
Mr. Neugebauer. As a follow-up to Dr. Reed, what areas or
what groups are not able to access existing technologies today
because there is not the appropriate infrastructure in place
to--for them to be able to access it? Who--what groups would
you say?
Dr. Reed. Well, we provide access, as I said, essentially
to anyone. In fact, one of the hallmarks of the NSF Centers has
been that we provide access to researchers regardless of the
source of their funding. So researchers who have DOE awards,
National Institutes of Health awards, NASA awards, all kinds of
people have gained access to the machines over the years via
peer review. The thing that is necessary for certain groups to
gain access is, if I can use a--perhaps a bad analogy. Here is
an old cartoon that shows a monster at the edge of the city
with a big sign that says, ``You must be at least this tall to
attack the city.'' In order to be able to effectively use high-
end computing systems, the application group needs to have
sufficient support infrastructure in terms of software
development, testing, being able to manage the results, and
correlate those with experiments. So it is possible for single
investigator researchers to use facilities, but it is much more
common for an integrated group of people, perhaps five or ten
people, faculty, staff, research associates, to work together
to be able to manage a large scale scientific application.
In terms of other components that are necessary, clearly,
high-speed networks are the on-ramp for remote access. And as
Dr. Freeman noted earlier, one of the things that is
increasingly true is that it is not just computing but access
to large amounts of data, scientific data archives, from
instruments as well as computational data. So the last mile
problem, if you will, for universities is part of the issue,
having that high-speed capability to be able to move data back
and forth, as well as the local capability. The other thing I
would say, the most common model of the way machines are used
these days is that local researchers typically will have some
small copy of a machine on which they do local development and
testing. And then they will run it scale at the national
center, so that local infrastructure is an additional part of
the story.
Mr. Neugebauer. Thank you, Dr. Reed and Mr. Chairman.
Chairman Boehlert. Thank you very much.
The gentleman from Oregon, Mr. Wu.
Mr. Wu. Thank you very much, Mr. Chairman. And I don't have
so much a question right now as a comment. Mr. Chairman, just
as you shared a story earlier about the Nobel Prize winner who
is, you know, having access problems and all, I just recall a
quite memorable moment. This was probably about 10 years ago in
the mid-90's or so. And I was practicing law in the technology
field. A friend of mine took me on a tour of--he was a software
developer for Intel, and he took me on a tour. And he did
supercomputing software. He took me on a tour of their
supercomputing inventory, I guess you would call it. I didn't
see the production areas, but we were in a room maybe half the
size of this one, and it had a lot of HVAC in it and a bunch of
black boxes, scaleable black boxes. They were massively
parallel supercomputers.
And there was just this one moment when he paused and
said--I can't imitate his accent, but he said something to the
effect of, ``You know, David, you are in the presence of one of
humanity's great achievements, like the pyramids.'' And I will
just always--I will remember that. It was striking to me,
because Intel got out of the business a little while later. And
the concern, I guess, besides the recollection, the concern I
want to express is that, you know, civilizations and
technologies ebb and flow, and the pyramids are still there.
The civilization that created them stopped doing things like
that, and for the most part, is no longer around. We had a one
shot--well, not a one shot, but a limited series of shots of
the moon in the--our space program, there is a sense that there
is a high water mark there, and perhaps we have receded from
that. I have no sense that our computing industry is
necessarily at a high-water mark and receding.
But I just want to encourage you all and this Congress and
our government in general to support a sustained push in this
and other arenas so that we do not look back at some distant
future date and, like me, think back to that supercomputing
inventory where the business no longer exists. And you know,
people in business have to make business decisions. But as a
society, there is a necessary infrastructure where we need--
whether it is a space program, whether it is biology, whether
it is computer technology, we need a long, sustained push. And
it is not necessarily sexy. Those black boxes were not
particularly notable from the outside. And I just want to
encourage you all and encourage this Congress to sustain that
necessary long-term push for the long-term health of our
society.
And thank you. I yield back the balance of my time, Mr.
Chairman.
Chairman Boehlert. The Chair wishes to identify with the
distinguished gentleman's remarks. I think the Committee does.
The Chair recognizes the distinguished gentleman from
Georgia, Dr. Gingrey.
Dr. Gingrey. Thank you, Mr. Chairman.
Dr. Freeman, I am going to fuss at my staff, because they
didn't call to my attention that you were the founding Dean of
the College of Computing at the Georgia Institute of
Technology, my alma mater.
Dr. Freeman. Well, I will do the same. My staff didn't
alert me that you were a Georgia Tech graduate. But I
discovered that early this morning.
Dr. Gingrey. Had I known that, I would have taken the
liberty, as my colleague, Mr. Johnson, did earlier, in
introducing Dr. Reed. I have, actually, two children who are
also Georgia Tech graduates. Now I have to admit, when I was at
Georgia Tech, it was in the days of the slide rule. And if you
wanted to see a teraflop, you just went to a Georgia Tech gym
meet.
Dr. Freeman. It is probably still true.
Dr. Gingrey. And even today, I have to say that I still
think a laptop is a supercomputer. But the staff has done a
good job of feeding me a couple of questions. And I want to
direct the first one to you, Dr. Freeman.
You stated emphatically that the National Science
Foundation is committed to stable, protected funding for NSF-
supported Supercomputer Centers. Yet we hear from scientists
and from others in the Administration that NSF's focus in the
future will be on grid computing and putting high bandwidth
connections between the centers and not on supporting top-of-
the-line supercomputers at the centers. At what level do you
currently fund the Supercomputing Centers? Are you committed to
maintaining that level of support separate from the grid
computer effort over the next several years? And does the
National Science Foundation plan to support future purchases of
new supercomputers at the centers?
Dr. Freeman. Let me note, Dr. Gingrey, that there are a
number of ways in which the Supercomputer Centers, specifically
San Diego, Illinois, headed by my colleague here, and at
Pittsburgh, there are a number of ways in which they are
supported. They have grants and contracts with a number of
agencies as well as with a number of different parts of the
National Science Foundation. Specifically, the program that is
under my peer review has been called the PACI program, the
Partnerships for Advanced Computational Infrastructure. That is
a program that, for the two partnerships, one located at
Illinois under Dr. Reed's direction, one located at San Diego
under Dr. Fran Berman's direction, each of those partnerships
receives about $35 million a year under the current
arrangements.
Some of that money is then reallocated as it were by them
to some of those partners, other universities that either
operate machines or participate in the development of the
advanced software, both applications and the underlying
enabling software. In line with our recommendations of our
distinguished panel, the Atkins Committee, we are in the
process of restructuring the way in which that support is
provided. We have, in no way, indicated that the total amount
of money going to those activities would decrease. In fact, in
speeches both to Dr. Berman and Dr. Reed's partnerships groups,
I have indicated our intent to do everything possible to see
that the amount of money going into the activities in which
they are participating would grow over time.
Dr. Gingrey. And a follow-up question, too, Dr. Freeman.
The testimony today has emphasized that high performance
computers are not just about hardware and big machines. Dr.
Reed and Dr. Orbach both described the importance of investing
in research on supercomputer architecture and on the algorithms
and software needed to run the computers more effectively. What
agencies support this sort of research? Does the industry do or
support this sort of research? And is NSF responsible for
supporting fundamental computer science research in these
areas?
Dr. Freeman. The--your last statement is the core
characterization of the directorate that I head. Our budget of
about $600 million a year, in round numbers, current year.
About $400 million of that goes largely to the support of
fundamental computer science and computer engineering research
in this country. Overall, that supports slightly more than half
of the fundamental computer science research in this country.
Let me come back to the opening part of your last question
there and emphasize, and I had wanted to in response to one or
two earlier questions, I understand that today's hearing is
focused on high-end computing, on supercomputers, as it should
be. But I believe it is extremely important to keep in mind
that there must be a balance, one, and two, that without
networking, without software, there is no access. The Japanese
supercomputer is, in fact, a case in point. It may be the
fastest machine on the planet at the moment, at least in the
public domain that we can speak about, but it is not connected
to the Internet. So if someone in your district wants to go and
use that computer, they would have to travel to Japan.
Secondly, it basically doesn't have much software. And many
of the inquiries we have been getting from the scientific
community is--are of the form, ``How can we put some of the
software that we know how to build in this country, how can we
put that on that Japanese Earth Simulator machine?'' So I would
use a simple analogy. A Ferrari is a very fast car, but you
wouldn't use the Ferrari to transport oil in or to take honey
or to do a lot of other things in. So having just high speed is
not the only thing.
Dr. Gingrey. Thank you, Dr. Freeman. And thank you, Mr.
Chairman.
Mr. Ehlers. [Presiding.] The gentleman's time is expired.
Next, we are pleased to recognize the gentlewoman from
California, Ms. Lofgren.
Ms. Lofgren. Thank you, Mr. Chairman.
First let me apologize for my delay in coming to this
hearing. The Judiciary Committee was marking up the permanent
Internet tax moratorium bill, and we successfully passed that,
so I think that is good news for the high-tech world. But I do
think this hearing is extremely important, and I am glad that
the hearing has been held.
Like my colleague from Oregon, Mr. Wu, I remember being at
Genentech a number of years ago, and with then Vice President
Gore. And there were so many smart people, and they had Ph.Ds
in various biological fields. And the Vice President asked them
what was the most important technology for their work with the
human genome, and they spoke with one voice, ``Computers.'' And
really, without an adequate investment in high-end computing,
we will see a shortfall in the innovation that we need to have
in a whole variety of fields. So this--including fusion, I was
glad to hear that the Chairman had identified that.
I have really just two quick questions. The first may be a
stretch, but we have, in a bipartisan way, been attempting to
remove the MTOP measurement as a restriction for exports of
supercomputers. And the question really is, and I don't know
who is best suited to answer it, whether that misplaced
outdated measurement might be impairing the private sector
development of high-end computing, because it is limiting
markets. That is the first question. And the second question,
which probably goes to Dr. Freeman, is whether we are
sufficiently investing in materials science that will, in the
long run, form the basis for next-generation supercomputers. So
those are my two questions, and whoever can answer them, I
would be delighted to----
Dr. Freeman. I am afraid I don't have--I am not sure any of
us have much information on the export issue. I was----
Ms. Lofgren. Okay. Fair enough.
Dr. Freeman. I must leave it at that. As to the second--
your second question was concerning the investment in materials
science. Again, that is not my particular area of expertise. I
would note that the nanoscience, nanotechnology initiative of
the government that NSF, I believe, is leading and that DOE is
heavily engaged in, certainly seems to be pushing the
boundaries there. The specific question, I am afraid, I don't
have any information on it.
Dr. Orbach. I could add that the material science area is
essential for computation development, especially for the new
architectures. And here, DARPA plays a very key role in
experimenting with new architectures, new structures that will
lead to advanced computing efforts. And most of them are
controlled by material science issues. This is an area that all
of us are investing in heavily.
Ms. Lofgren. All right. Thank you very much.
Dr. Reed. I could, perhaps, provide just a bit of insight
into your MTOP question. It does and it doesn't. One of the
things that has been true in parallel computing is that one can
export individual components and then often third parties will
reassemble them to create machines that are then sold
internationally, independently of the original vendors in the
U.S. The place where it is, perhaps, more of an issue is in
monolithic high-performance machines. The vector computers
where one, by very obvious means, can't disassemble them and
reassemble them elsewhere. And it does have an impact there,
but it is a mixed impact.
Ms. Lofgren. Thank you. I yield back, Mr. Chairman.
Mr. Ehlers. Well, will you yield, just a moment, to me?
Ms. Lofgren. Certainly.
Mr. Ehlers. Let me follow up on that. The Japanese are
not--who are the leaders in vector machines, don't want to have
any such restrictions, is that correct?
Dr. Reed. That is correct.
Mr. Ehlers. So doesn't that render our restrictions
meaningless?
Dr. Reed. It does, in many ways. In fact, those Japanese
machines are sold in the markets, yes, where the U.S. vendors
cannot sell.
Mr. Ehlers. Yeah.
Ms. Lofgren. Mr. Chairman, I wonder, I don't want to
belabor the point, but whether we might consider having some
further deliberations on this MTOP issue in the Science
Committee, because I think we have a unique perspective to--it
is not really the main point of this hearing. But I think we
could help the Congress come to grips with this in a way that
is thoughtful if we did so, and I----
Mr. Ehlers. Well, if you will yield.
Ms. Lofgren. Yes.
Mr. Ehlers. Obviously, I don't control the agenda of this--
--
Ms. Lofgren. Right.
Mr. Ehlers [continuing]. Committee, but I would say this is
just a continuing, ongoing battle. And I think you and I are
both engaged in the----
Ms. Lofgren. Right.
Mr. Ehlers [continuing]. Encryption battle, which--in which
the net effect of the controls was to strongly encourage the
encryption industry outside of the United States.
Ms. Lofgren. Correct.
Mr. Ehlers. And it would, frankly, hurt our country. And I
don't know--I fought that battle very hard, and I was astounded
at how difficult it is to----
Ms. Lofgren. Right.
Mr. Ehlers [continuing]. Win those battles for people and
use national security as the excuse for their suspicions and--
or support for their suspicions, and it would be the detriment
of our country.
Ms. Lofgren. I--my time is expired, and I would associate
myself with the gentleman's remarks.
Mr. Ehlers. The gentlelady's time is expired.
And next, we turn to Dr. Bartlett, who has returned.
Dr. Bartlett. Thank you very much. I am sorry I had to
leave for an appointment for a few minutes.
In another life about a third of a century ago, I worked
for IBM. And we at IBM were concerned that as a company and as
a part of this country that we, as IBM in this country, were at
risk of losing our superiority in computers to the Japanese.
Apparently our worst fears have been realized. And our
reasoning for that, and this was a time when we were really
premier in the world, was that every year the Japanese were
turning out more, and on average, better scientists,
mathematicians, and engineers than we were. And we just didn't
think that we, at IBM, and we, the United States, could
maintain our superiority in computers if this trend continues.
The Japanese now have, as has been noted by our Chairman in
his opening statement and by others, the Japanese now have the
world's fastest and most efficient computer. How did they do
it? Was it people or was it something else?
Dr. Freeman. Let me start the responses. First let me note
that I believe that while the Japanese may have the fastest
computer, that it would be erroneous to say that they have the
complete lead in computing, because there are a number of other
elements that are extremely important that they do not have the
lead in. And indeed, as I noted a moment ago, I believe you
were out of the room, in order to utilize that Japanese
machine, they are fairly urgent to get some of our American
software technology to enable the usage of that machine. I
would underscore what Dr. Reed said earlier this morning, and
that is that it is not that the Japanese were smarter. Indeed,
a number of the engineers, I understand, that worked on the
design of that machine, in fact, had spent time in this country
as employees. And I know people who know them and don't think
they are particularly any geniuses, obviously very good
engineers, but that it has been the sustained focused
investment that has been made and that is, if anything, the
single key reason why they now have a Ferrari and we may not
have.
Dr. Bartlett. Dr. Orbach.
Dr. Orbach. I would like to congratulate the Japanese on
their achievement. What they did was to build a balanced
machine. Instead of going for just the highest speed they
could, they recognized the importance of balance, the
communication with the memory, the speed of communication
between the microprocessors and the memory. They put together a
machine that would be suitable for scientific, and actually, I
believe, industrial uses. It wasn't just a peak speed machine.
It was a beautiful device.
That is what we are trying to achieve ourselves now. We
have learned from the Japanese that it can be done. We are
experimenting with different architectures, not necessarily the
same, to create something balanced, something that would
function for problem solution as opposed to just some arbitrary
speed criteria.
Dr. Reed. I would echo that. The sustained investment is
critical. I go back to what Edison said years ago that genius
is 1-percent inspiration, 99-percent perspiration. The
sustained effort is really what has been critical. And that, I
think, is really the lesson for us, that we need to look at the
sustained level investment. The broad array of capabilities and
computing technology in the U.S., I think it is absolutely
true, as Dr. Freeman said, that overall the preponderance of
capability in the world remains in the U.S. We have not
harnessed it as effectively as we could, and the Japanese is
a--machine is a great example of what can happen when the
appropriate talent and sustained investment is harnessed.
Mr. Scarafino. One of the other items that, I think, had an
effect on direction was the way that metrics are used, the way
that we compute what we think that the supercomputer is. They
really are based on what individual processors can do running
specific benchmarks. Microprocessors got very fast. I mean,
Moore's Law basically allows them to be faster and faster as
there--as the new generations come out. But they don't address
how the overall system performs. If you look at the way that
the top 500 supercomputer list is, it takes what effectively
any one component of a massively parallel machine can do and
multiplies it by the number of elements that are in there. And
that is a theoretical peak number, but it is something that, in
a practical application, can never really be achieved.
And I think that is something they are understanding better
now that we have built some of these large machines and now
that we have tried to run actual applications on them where we
are seeing five percent of peak available. Even with a lot of
work in order to get up to those high numbers, we are starting
to understand that the metrics that we are using to describe
these machines lack some of the characteristics, that they
don't really reflect the actual reality of the situation. If
you look at the number that the Earth Simulator has with regard
to the metrics versus the next fastest machine, I believe the
actual reality of the disparity between the two machines is
much greater than those numbers show.
Mr. Ehlers. The gentleman's time is expired.
We are pleased to recognize the distinguished gentleman
from Tennessee, Mr. Davis.
Mr. Davis. Thank you, Mr. Chairman. It is good to be here,
and I apologize for being late when most of you gave your
testimony, so therefore, I have basically reviewed some of the
testimony that has been--that was given while I was at another
meeting. It seems like that, here in the Congress, we have four
or five meetings at once. I understand the term being at two
places at once. Sometimes maybe it is four or five.
Dr. Orbach, I have a couple questions for you. One will
deal, obviously, with the supercomputers that--in the different
labs, and the one I am most interested, obviously, is the one
in Oak Ridge, which is part of the district that I represent.
But as I have read your testimony, you mentioned that there is
a new sociology that we will need to develop when it comes to
supercomputers, a sociology of high-end computation that will
probably have to change. What--exactly what do you mean by that
when we talk about the new sociology and how we may have to
change?
Dr. Orbach. The size of these computers and their speed is
beyond anything that we have really worked with before. And it
is my view that it will take teams of scientists to be able to
utilize these facilities in their most efficient fashion. And
not just the scientists interested in the solution of the
particular problem, but applied mathematicians, computer
scientists, people who can put together a team to actually
utilize these machines. It is my view that these machines will
be like light sources or like high energy physics accelerators.
Teams will come to the machine and utilize it as users. And by
the way, industry, also, will come and use these machines as a
user group.
We don't do that now. And when I made earlier reference to
opening NERSC, our big machine, up for grand challenge
calculations, part of that was directed toward this new
sociology. How can we learn how to use these machines most
effectively? The amount of data that they can produce is huge.
Our ability to understand it in real time is limited, that we
need visualization methods to understand what it is we are
doing. There is so much coming out. These are things that we
really have to get used to learning how to do. And even the
Japanese themselves on the Earth Simulator are finding that it
is difficult to understand, in real time, what you are doing.
And so they are going to visualization methods. It is that that
I was referring to.
Mr. Davis. Okay. It--when the Earth Simulator first came--
obviously most scientists, I think, were aware that Japan was
working in this area, somewhat surprised, I think, when it came
on line as quickly as it did and it came up as quickly as it
did. I know that there are locations throughout America that
basically got excited as well in this direction. ``We have to
be competitive. We have to find or develop a supercomputer or
at least the area for it.'' I--again, I am selfish, so I think
our folks, in the area where I am from, at the--which is one of
the labs that does a great deal of work. We have got the SNS
[Spallation Neutron Source] project there, which is a part of--
when you talk about sociology, there are a lot of folks that
are visiting scientists. And my hope is that as you start
looking at locations, I know there is, what, a $35 or $40
million increase for supercomputers in the budget and the House
is looking at the same thing as the Senate. I don't even know
what kind of criteria that you will use as you determine the
location: maybe the capabilities, capacity, some of those
issues as you look at how you will make that decision of where
those dollars will be spent and the lead location of--for our
supercomputers.
Dr. Orbach. We anticipate that there will be more than one
lead location. Our investment now is to look at new
architectures. The X-1, for example, at Oak Ridge National
Laboratory, is but one example. As has been discussed before,
the number of vendors in the field is too small. We are
encouraging industry to develop interest in high-end
computation. As these new vendors come on line and decide to
get interested in this, we could see a multi-center development
of--and testing of these new facilities. Ultimately, when the
big machines are built, and I hope there will be more than one,
there will be a competition between the laboratories as to
which would be the most effective place for the computation
facility to be housed.
Mr. Davis. Of course maybe I shouldn't press them further,
but we have got cheap electricity--thank you so much.
Do I have more time? I have one more question to Mr.
Scarafino.
Mr. Smith of Michigan. [Presiding.] You don't. Your time is
up, but maybe if it is a short question----
Mr. Davis. It would be. Mr. Scarafino, I know that the
automobile industry, especially from some of our competitors in
Asia, automobiles may be--I don't want to say quieter than the
American vehicles, because I drive an American vehicle,
probably always will. But are you able to work--I mean, if--do
you sense that industry in this country has the connection, the
ability that supercomputers are available to you or will be in
the future? And would you use this asset in private industry if
that was made available?
Mr. Scarafino. You made a reference to vehicles being
quieter. Is that what----
Mr. Davis. I am just saying, when you look at the auto
industry in Asia, it seems that their cars--I am getting in
trouble with you, aren't I?
Mr. Scarafino. No. No. It is a reality and----
Mr. Davis. They may be quieter. And some folks think,
perhaps----
Mr. Scarafino. More durable, at least.
Mr. Davis. I don't know about more durable. I mean, I
drive--I have always driven American cars, and I don't see that
they are any more durable.
Mr. Scarafino. Okay.
Mr. Davis. I get 250,000 miles. Mine always rode finer
before I sold them, so--but it seems that industry in some of
the Asian countries that are our competitors are being allowed
or given the opportunity to use the assets that may be funded
by government more so than private industry. Is that the case,
and should that be changed?
Mr. Scarafino. Well, I don't know where the funding comes
from. I know that some of our challenges, one with regard to
the--to noise, for example, has to do with wind noise in
particular that we are running on the fastest machines we have
access to. And as I had stated before, some of these problems
run for two weeks on a computer to get a single answer. Toyota
does have a fair number of vector machines that is--but they
are either NEC machines, and they also have Fujitsu machines,
which are, you know, basically Japanese vector high-end
processors. I don't know exactly what they run on them, so I
mean--so it could be implied that they are utilizing these
machines in order to create the quieter vehicles they have. But
that is--it is kind of speculation on my part.
Mr. Smith of Michigan. The Chairman would call on the
Chairman of the Subcommittee on Space, Mr. Rohrabacher.
Mr. Rohrabacher. Thank you very much.
Maybe you can answer a question for me. I have just got
some very fundamental questions about computers. Way back when,
you know, like when I was a kid, I seem to remember--and that
was a long time ago. This is like after World War II. Weren't
there computers that--didn't they have cards that you punched
out, and was it an ``x'' and a ``y'' or something like that or
a ``y'' and a ``0''? Or what was it? Do any of you remember
what I am talking about? Is that what it is still based on?
Does it still go back to that fundamental principle whether two
things being punched out on a card and we have built from
there?
Dr. Orbach. Well, the--as I am older than you are, and I,
believe me, carried around those crates of punched cards from
one machine to another. They--if it was punched, it was a
``1''. If it wasn't punched, it was a ``0''. That digital----
Mr. Rohrabacher. Okay. ``1'' and ``0''.
Dr. Orbach. That digital structure still is underpinning
our computational methods.
Mr. Rohrabacher. All--so everything we are talking about
still goes back to that ``1'' and that ``0''?
Dr. Orbach. That ``1'' and that ``0''. Now there are
quantum computation methods that are being explored that will
bring an additional richness to that. These machines--NSF is
supporting research in that area. DARPA especially is
supporting research on some of these advanced concepts that
will carry us beyond the ``0'' to ``1'', the binary methods
that we have been using over the years. And the richness of
some of these methods is very exciting. It will be interesting
to see how they develop.
Mr. Rohrabacher. So we are right now thinking about that
basic fundamental may change now and take us into something
that we--may even be more spectacular than what we have got?
Dr. Orbach. Yes.
Mr. Rohrabacher. Let us think about that. I also remember
when I--of course we are talking--I remember people talking
about a lot of everything that we were doing in computers was
based on sand, and making glass out of sand and electricity
being run through sand that was reconfigured. Is that still the
case?
Dr. Orbach. Well, I think they were referring to the
silicon. Silicon dioxide is sand.
Mr. Rohrabacher. Right.
Dr. Orbach. But yes, the silicon-based machines, microchips
are still the base of choice, though there are other
semiconductors that are used. But it is very plentiful, but it
has to be purified to a degree that makes it tricky and
expensive.
Mr. Rohrabacher. We have turned this ``1'' and ``0'' and
sand into enormous amounts of work and wealth creation. That is
mind boggling. I used to be a speechwriter for Ronald Reagan
back in the 80's. And that is getting afar back. Now I was
asked to go to the Soviet Union, back when it was the Soviet
Union, right after Gorbachev took over and things were
beginning to fall. And something really--I had a little
experience there that really taught me--and we, because there
had been all of this talk about supercomputers and--at that
time. The supercomputer was going to change everything. But
when I went over to Russia, I took with me a bottle of peanut
butter, because I knew they couldn't make peanut butter. At the
right moment, I had a bunch of college kids talking to me about
America. I pulled out the jar of peanut butter, and I--you
know, you can imagine if you have never tasted peanut butter.
But getting on with the story, one of them came up to me
after hovering with the other kids there and said, ``What are
the black marks on the side of the peanut butter jar?'' I said,
``Well, that is a bar code. And that is where every time I go
to the store and buy a jar of peanut butter at the food store,
it itemizes my bill for me. The computer itemizes the bill, and
an inventory is notified that there is an item that has been
sold. And you know, that makes everything easier.'' And the
kids got together. These are college kids in Russia at the
time. ``You know, that is why we don't trust you Americans. You
know, you are lying to us all of the time. Computers at a food
store? Give me a break.'' And they could not believe that we
had--and I went to their food store and of course they were
using abacuses and things like that. And it is--you know, that
impressed me.
That is when I knew we were going to win the Cold War. I
mean, there was no doubt about that. It wasn't just the peanut
butter. It was the fact that we had computers being put to use
across the board in our economy and food stores where they
couldn't even think about doing it at food stores.
Now today, when we are--I am very happy that Mr.--for Dr.
Freeman and Dr. Orbach have talked about balance. Are we
balanced in the fact that we are working to make sure that
there is a widespread benefit to this sand and computer-
generated numbers? A widespread benefit versus only putting our
eggs and trying to create a pyramid, so to speak?
Dr. Orbach. If I could respond, I hope we are. We are
certainly encouraging industry, as you described it, to join us
in this quest for seeing what can be accomplished at these
speeds. Both NSF and DOE and actually the Office of Science and
Technology Policy have included industry in the development of
these new machines for exactly the purposes that you recognized
before. And as you have heard from the representative from
Ford, but also from GE and GM and other companies, there is
widespread recognition that these machines will give us
economic competitiveness, exactly as you described, that nobody
else can get.
Mr. Rohrabacher. Across the board rather than just----
Dr. Orbach. Across the board.
Mr. Rohrabacher [continuing]. At the huge level?
Dr. Orbach. We won't have to build models. Do you remember
the wind tunnels with airplanes and they put them in? If we
have enough speed, we can model that computationally and do it
in a matter of hours or weeks, at the most, instead of years.
We can save huge amounts of monies on these, what we call,
virtual prototypes that will free us from having to build
specific models.
Mr. Rohrabacher. And then we could spend that money
elsewhere on something else?
Mr. Smith of Michigan. Yeah, I was going to ask Mr.
Scarafino if--how many crash dummies the Cray computer has
saved, but you--are you----
Mr. Rohrabacher. I guess I am done. Let me just note that
Tim Johnson wanted me to, for sure, say that--to Professor Reed
that your institution has such a great record of achievement
and wanted me to ask one question that is--and this is the
easiest question you will ever get from Dana Rohrabacher, just
what do you attribute your great record to?
Dr. Reed. You are right. That is an easy question. Well,
let me hop back to give you a serious answer, because I think
it is important. I mean, the success of the high-end computing
infrastructure has really rested on a combination of the fact
that there is a high-end infrastructure but also on the
critical mass of people. You bring great people together to
work with the world class infrastructure and on really critical
applications, and interesting things happen. And that is where
the web browser spun out of NCSA. It is where a lot of the
technologies that will enable cyberinfrastructure have spun
out, things like large-scale data mining. There are lots of
collaborations with major industrial partners that, as Dr.
Orbach said, it saved major corporations lots of money by
allowing them to think smarter, to avoid killing crash test
dummies, by applying technology in creative ways. And so what
we look for are opportunities where the combination of people
and technology can apply.
You mentioned the broad spread applicability of computing.
The next big wave beyond, sort of, the grocery store and
personal computer kind of thing is--computing is really
becoming ubiquitous. One question I often ask people is: ``How
many computers do you own?'' And if they have kids, the answer
might be three or four. But if you bought a car in the last 20
years, you have anti-lock brakes, you have an electronic
thermostat. The answer is really hundreds. And that
proliferation of technology, the ultimate success is the extent
to which it is invisible and it enriches and empowers people's
lives. And that is the broad tie that really is where things
like NCSA and other centers have had an impact.
Mr. Smith of Michigan. Gentlemen, you have been very
patient today. Thank you for your time and patience. You only
have one more questioner to face.
Let me--I want to get into the competitiveness and the
economic stimulus to the country that might have this
computing. Let me start out with the Earth Simulator. Do--does
Japan sell time on that to our scientists in this country?
Dr. Orbach. The Japanese have made time available. Mr.
Sato, who is the director of that machine, has about 16 to 20
percent of the machine available to him. And he has been very--
he spent time in the United States, in fact, studying science,
computer science here. He has made the machine available to our
scientists gratis. The somewhat worrisome feature is that our
scientists have to go to Yokohama to use it, and the
discoveries are going to be made there. That is what we are
worried about. One of the interesting structures, again, the
sociology of how the machine is being used, is that there are a
number of Japanese young scientists who are there while our
scientists are developing their codes and using that machine.
So they get the advantage of discovery on the site. And as you
have already heard, it can't be networked. They did that on
purpose. So they have the availability of discovery, but we
don't. And----
Mr. Smith of Michigan. Well, at least a little more access
and a little more convenience. Dr. Reed, somebody, help me
understand a little bit. Our--and Dr. Orbach, it is sort of
tied in with--I mean, with--has IBM and Argonne and certainly
the Berkeley National Lab developed new computer systems that
are going to be even faster than the Japanese current model?
And with the so-called development of the--a Blue Gene
computer, and I--where we go with that computing capability,
which is much faster, how does that tie in terms of our
ability? Is it a concern of--the first one, I guess the
estimated development time is '05 some time and maybe later for
the other one. How does that tie in to the needs of this
country and the competitive position in terms of high-end
computers?
Dr. Orbach. It is one of the architectures that I mentioned
before that we are exploring. We are looking at different
architectures to see how efficient they are and how useful they
are for particular scientific problems. And indeed, the Blue
Gene, with the collaboration you have made reference to, is a
very interesting and important component. It is one of four or
five different architectural structures that we are looking at.
Dr. Reed. It is certainly the case that the Supercomputing
Centers that NSF funds have pretty deep collaborations and
connections with the DOE efforts. Argonne National Lab that you
mentioned is in Illinois. And we have a jointly funded effort
with machines deployed at Argonne and at the University of
Illinois. And we are also involved in collaborations with them
and with NERSC looking at applications of that machine and what
architectural features will be met--best matched to
allocations. We certainly hope that in the coming years we will
have the resources to be able to deploy one of those machines.
Mr. Smith of Michigan. Well, it--as I think you know, I
chair the Research Subcommittee, that has oversight over the
NSF. And Dr. Orbach, as--if it continues to develop that Energy
plays a more prominent role in the development of high-end
computers, NSF has been very organized to allow a lot of use in
our labs. Is Energy going to be in a position? You mentioned 10
percent of potential usage now, but it is going to take not
only--it is going to take much more than that if you are going
to accommodate the demand for that kind of time. And it is
going to take both organization and facilities, I would think.
Dr. Orbach. I think you are exactly right. We will--we
depend on the National Science Foundation network for coupling
to our large-scale computers. And we work very closely with
them. Our initiative in high-end computation will be open to
everybody. Anyone who can--who wishes can compete for time,
subject to peer review, on these new high-end machines. And so
we will--we are working hand in glove with NSF to make them
accessible. The challenge now is to determine which
architecture we want to invest in and create the structure that
would enable our scientists to perform their needed
calculations.
Mr. Smith of Michigan. In--like our basic research that is
oft times sophisticated into application in other countries
quicker than we do it in this country with our mandate for
publishing any time federal money goes into it, likewise give
me your impression as we face a more competitive economic
market throughout the world with everybody else trying to
produce our cars and everything else as efficiently as we do
and as quality conscious as we have been. With our allowance of
scientists from other countries to use our--whatever computers
we develop, how are we going to still have the edge in terms of
the application in this country of the greater efficiencies and
development of new products and better ways to produce those
products? How are we going to make sure that our investment has
the advantage in this country?
Dr. Freeman. Well, let me just respond. I think certainly
in the general case, open science is still, by far, the best.
The competitiveness that the United States--the competitive
edge that the United States has, in many, many cases, is due
specifically to very talented, very bright scientists and
engineers who have chosen to come to this country to study at
our universities and oftentimes stay so that open science----
Mr. Smith of Michigan. Well, except in--that is changing
very rapidly. They oftentimes choose to stay, but it is now a
lesser option since 9/11, and so that should concern us.
Dr. Freeman. In some--I quite agree that that is a concern,
but as I noted in my opening remarks, the synergy that has
existed between industry and basic research, for example, has
been very productive. And we certainly want to see that
continue.
Mr. Smith of Michigan. Can I get your reaction from Ford
Motor Company, Mr. Scarafino?
Mr. Scarafino. I am in agreement. What we really need is
access to the kind of computing capability that we can give to
our engineers. And by giving them those kind of tools, I am
very comfortable that they will know exactly how to make the
best use of them. And we would remain--provide us with a way of
being competitive, even if those tools are available other--in
other places in this world.
Mr. Smith of Michigan. Well, I did--let me wrap this up by
saying, Dr. Freeman, it still concerns me when you said
students from other countries come in and often stay here.
Right now half of our research through NSF, for example, is
done by foreign students. As we try to do what we can to
stimulate interest and stimulation for American students to
pursue math and science careers, the question earlier was asked
who is--why is Japan developing these kind of high-end
supercomputers instead of the United States. Can we expect in
the future to see this kind of development, both of hardware
and software, go to countries that give greater emphasis to
math and science, whether it is China or whether it is India or
some other country that tends to encourage and push students in
that direction? Is--do you see that as the trend, Dr. Freeman?
Dr. Freeman. Well, yes. If I may respond there, I certainly
would agree with you that we need to get more Americans into
science and engineering. And as you well know, that is one of
the primary emphases of NSF across all fields. The computing
field is also certainly one of those where we are doing
everything we can to encourage more American students, not to
exclude the foreigners that choose to come here, because indeed
we need more people in general in the computing arena, but we
certainly must get more American citizens into the pipeline.
Mr. Smith of Michigan. Let us wrap this up. I--maybe if
each one of you have a comment of advice for the Science
Committee and for Congress in general in your arena, maybe if
you have between 45 seconds or so just to--any last thoughts
that you would like to pass on for the record for Congress and
the Science Committee.
Dr. Orbach. First, I would like to thank this committee
very much for holding this hearing. I think you have brought
out the key issues that we all are concerned with and we are
very grateful for having this hearing. I would like to comment
that on the competitiveness, American industry has always
stepped up to the plate. And they are a partner in the
development of high-end computation. And I think that is a very
special American trait. If we can keep our industry a partner
as we develop these new machines, I think it will show in the
marketplace.
Dr. Freeman. I would stress three things in closing:
supercomputing is important; but secondly, it must be looked at
and understood in the broader context that I believe all of us
have addressed this morning of storage, networking, et cetera;
and third, a very important topic that I believe has only been
brought up in the last few moments is that of education. If we
do not have the trained people, let alone to create such
capabilities, but to utilize such capabilities to make sure
that they are applied to, whether it is industry or the most
advanced most basic research, if we do not have those people,
then it makes no difference what type of supercomputers we
have.
Mr. Smith of Michigan. Dr. Reed.
Dr. Reed. I think we are on the cusp of something truly
amazing in what we can do with computational capabilities, some
very fundamental questions that are as old as mankind are
really close to being within our grasp. I think in order to
capitalize on that, we have to look at the level of sustained
commitment to build the kind of machines that will tackle those
kinds of problems. I think that is going to require coordinated
investment and activity R&D across the agencies. And I urge
that, you know, we look carefully at how to make sure that
happens so we can capitalize on the opportunity and maintain
the kind of competitive edge that we have historically had.
Mr. Scarafino. I would like to thank you for having this
hearing. It is good to know that the country is understanding
the importance of this area and is willing to basically make
more progress in it.
Mr. Smith of Michigan. Gentlemen, again, thank you for your
time and consideration and your patience. With that, the
Committee is adjourned.
[Whereupon, at 12:31 p.m., the Committee was adjourned.]
Appendix 1:
----------
Answers to Post-Hearing Questions
Answers to Post-Hearing Questions
Responses by Raymond L. Orbach, Director, Office of Science, Department
of Energy
Questions submitted by Chairman Sherwood Boehlert
Q1a. At the hearing you spoke of developing and purchasing new
supercomputers to be installed at Department of Energy (DOE) labs and
of making these computers broadly available to U.S. researchers. When
would such computers be installed and be open for general use?
A1a. Our IBM system at NERSC at Lawrence Berkeley National Laboratory
is currently open for general use--and has recently been upgraded to
over 6,000 processors, making it one of the largest machines available
for open science. In fact, we are setting aside 10 percent of this
resource for large problems with the potential for high scientific
impact. All researchers, regardless of their source of funding may
apply.
We are currently evaluating a small Cray X1 system at the Center
for Computational Sciences at Oak Ridge National Laboratory. As we
transition from evaluation and research into broader use, this system
will also be available. The initial evaluation is being done according
to an open plan and involves many researchers from the general
community already. Given the availability of additional funding, more
capable systems could be made accessible as early as FY 2005.
The Pacific Northwest National Laboratory recently announced that a
new Hewlett-Packard supercomputer with nearly 2,000 processors has been
installed in the Environmental Molecular Sciences Laboratory (EMSL) and
is now available to users. As a National Scientific User Facility, the
resources within the EMSL are available to the general scientific
community to conduct research in the environmental molecular sciences
and other significant areas.
Q1b. On these machines, would a certain percentage of the time be set
aside for scientists associated with DOE? What priorities would
determine who received what amount of time? What peer review mechanisms
would be used to make awards?
A1b. A percentage of time on these machines would be set aside for
scientists associated with DOE. We have procedures in place to allocate
time on these resources--to both DOE and non-DOE scientists. The
process is described below.
We have an Office of Science allocation plan in place. It allocates
time to our Associate Directors, who in turn allocate it to researchers
who are working on their science programs.
We have a peer review mechanism currently in place at NERSC. It has
withstood the tests of time and we plan to continue to use it as long
as it serves the community well.
All Principal Investigators funded by the Office of Science are
eligible to apply for an allocation of NERSC resources. In addition,
researchers who aren't directly supported by DOE SC but whose projects
are relevant to the mission of the Office of Science may apply to use
NERSC.
Four types of awards will be made in FY 2004.
1. Innovative and Novel Computational Impact on Theory and
Experiment--INCITE Awards:
Ten percent of the NERSC resources have been reserved for a new
Office of Science program entitled Innovative and Novel Computational
Impact on Theory and Experiment (INCITE), which will award a total of
4.5 million processor hours and 100 terabytes of mass storage on the
systems described at http://www.nersc.gov/. The program seeks
computationally intensive research projects of large scale, with no
requirement of current Department of Energy sponsorship, that can make
high-impact scientific advances through the use of a large allocation
of computer time and data storage at the NERSC facility. A small number
of large awards is anticipated.
Successful proposals will describe high-impact scientific research
in terms suitable for peer review in the area of research and also
appropriate for general scientific review comparing them with proposals
in other disciplines. Applicants must also present evidence that they
can make effective use of a major fraction of the 6,656 processors of
the main high performance computing facility at NERSC. Applicant codes
must be demonstrably ready to run in a massively parallel manner on
that computer (an IBM system).
Principal investigators engaged in scientific research with the
intent to publish results in the open peer-reviewed literature are
eligible. This program specifically encourages proposals from
universities and other research institutions without any requirement of
current sponsorship by the Office of Science of the Department of
Energy, which sponsors the NERSC Center.
2. DOE Base Awards:
Sixty percent of the NERSC resources that are allocated by DOE go
to projects in the DOE Office of Science Base Research Program. DOE
Base awards are made by the Office of Science Program Managers and
DOE's Supercomputing Allocations Committee.
The largest production awards are called Class A awards. These
projects are each awarded three percent or more of NERSC resources;
collectively they receive about 50 percent of the resources. In
addition, they may receive extra support from NERSC, such as special
visualization or consulting services.
Class B DOE Base projects are awarded between 0.1 percent and three
percent of NERSC resources.
All DOE Base requests are reviewed by DOE's Computational Review
Panel (CORP), which consists of computational scientists, computer
scientists, applied mathematicians and NERSC staff. The CORP provides
computational ratings to the DOE Program Managers on the computational
approach, optimization, scalability, and communications characteristics
of their codes. They rate how well the code description questions have
been answered.
Projects are rated on a scale of one to five.
3. SciDAC Awards:
Twenty percent of the NERSC resources that are allocated by DOE go
to Scientific Discovery through Advanced Computing (SciDAC) projects,
which are a set of coordinated investments across all Office of Science
mission areas with the goal of achieving breakthrough scientific
advances via computer simulation that were impossible using theoretical
or laboratory studies alone. SciDAC awards are made by the SciDAC
Resource Allocation Management Team. SciDAC requests are not reviewed
by the CORP.
4. Startup Awards:
Less than one percent of the NERSC resources are awarded to
principal investigators who wish to investigate using NERSC resources
for new projects. For FY 2004, the maximum startup awards are 20,000
processor hours and 5,000 Storage Resource Units. A request for a
startup repository can be made at any time during the year; decisions
for startup requests are made within a month of submission by NERSC
staff and can last up to 18 months.
Q1c. Before the new supercomputers are installed, which DOE facilities
(and what percentage of time on these facilities) will be available to
researchers not receiving DOE grants or not working on problems
directly related to DOE missions?
A1c. Currently, 10 percent of the time at NERSC is open to all
researchers. The Office of Science program offices are free to go
beyond that number when they allocate their portions of the resources:
NERSC is currently the only production computing facility funded by
ASCR. Should additional production resources become available, they
will be allocated in a similar fashion. Time will also be made
available on our evaluation machines, in a manner consistent with the
accomplishment of the evaluation work.
Because the new Hewlett-Packard supercomputer at PNNL is located
within the EMSL, and the EMSL is a National Scientific User Facility,
the new supercomputer is available to the general scientific community.
Although non-DOE funded investigators may apply for time on the new
system, applications must be relevant to the environmental problems and
research needs of DOE and the Nation.
Q1d. Are you working with the National Science Foundation to ensure
that scientists continually have access to a full range of high-
performance computing capabilities?
A1d. Yes. We have formal interactions with NSF (and other agencies)
through the National Coordination Office under the auspices of OSTP,
and numerous informal interactions in our planning and review process.
Q2. What is the level of demand for cycle time on all of the DOE
Office of Science supercomputers? Which scientists and applications are
users of the most time? How will you continue to support the DOE
scientific communities and priorities while opening up machines to more
general use?
A2. The level of demand for cycles on Office of Science computers is
growing exponentially--increasing by an order of magnitude about every
2.5 to 3 years. We are currently operating at full capacity, so
managing our out-year demand for additional cycles will be challenging.
Our biggest users include those scientists engaged in research on:
Accelerator design for high energy physics;
Quantum chromodynamics;
Fusion energy;
Climate simulations;
Supernova simulations;
Materials science; and
Nuclear physics.
Scientists in these research areas often use hundreds--and
occasionally use thousands--of processors and use weeks to months of
resource time.
Continuing to support the DOE scientific communities and priorities
while opening up machines to more general use will be a challenge. We
are committed to staying this course--because our open science research
depends on it. We will ensure that the resources are put to those uses
which uniquely exploit them--and do the most to advance the frontiers
of science.
Answers to Post-Hearing Questions
Responses by Peter A. Freeman, Assistant Director, Computer and
Information Science and Engineering Directorate, National
Science Foundation
Questions submitted by Chairman Sherwood Boehlert
Q1. Witnesses at the hearing and staff at the Office of Science and
Technology Policy have indicated that the National Science Foundation
(NSF) may be underfunding research on new architectures for high-
performance computing. How much is NSF spending in this area? How did
you determine this spending level? How does this level of funding
relate to your assessment of the need for research in this area? Are
any other agencies or companies funding this sort of research? To what
extent are NSF's programs coordinated with these other activities? If
NSF does not invest more in this area, are other agencies or private
entities likely to fill the gap?
A1. NSF funding impacting new architectures for high-performance
computing can be tallied in several ways. In the narrowest sense,
funding that may lead most directly to new computer processors is
estimated to be around $5 million in FY 2003. Funding at this level is
provided from the Computer Systems Architecture program.
In addition to specific research on computer systems architectures,
NSF supports research and education in other areas critical to progress
in high-end computing, including algorithms, software, and systems.
Algorithmic research is essential to enable the transformation of
traditional sequential algorithms into parallel ones, and to find
algorithms for new classes of problems and new types of architectures.
Research on compilers, operating systems, networking, software
development environments, and system tools, is also essential if high-
end computing is to succeed. It is difficult or impossible to separate
out which of this research is directly applicable to high-end
computing, but we estimate it to be at least $50M in FY03.
In the broadest sense, NSF funding of the Extensible Terascale
Facility can be viewed as research on a specific architecture for high-
performance computing. As detailed below,
$126M has been spent on this advanced R&D project to date. The high
level objective is to show the feasibility of providing high-
performance computing capability via a highly-interconnected,
distributed set of computational resources.
In addition to this well-defined project, NSF invests in higher-
risk, longer-term research to ensure that innovation is possible years
from now. Examples showing promise for high performance computing
applications include nanoscale science and engineering, quantum
computing, and bio-inspired computing. NSF investments in these areas
are in excess of $100 million.
As with all NSF investments, funding levels for computer systems
architecture-related activities have been and continue to be determined
by a combination of inputs from the communities using high-performance
computing in their research, communities with expertise in developing
architectures and their supporting technologies, interactions with
other agencies, inputs from Congress, and the funds that are available.
NSF's investments in high-end-computing-related research complement
investments in computer system architectures, algorithms, parallel
software, etc. being made by other federal agencies, especially DARPA,
DOD, DOE, and by industry. As in other fields of science and
engineering, NSF draws upon partnership efforts to leverage these
investments. Specifically in the area of high-end computing they are
coordinated through the High-End Computing program component area of
the Networking and Information Technology Research and Development
(NITRD) Working Group. This partnership has recently been strengthened
through the focused activities of the High-End Computing Revitalization
Task Force.
As part of the NITRD interagency coordination effort, NSF is
considered to have over $200M in ``high-end computing infrastructure
and applications'' and nearly $100M in ``high-end computing research
and development.''
Finally, NSF capitalizes upon the outcomes of Federal Government-
supported nearer-term high-end computing research and development
activities enabled by its sister agencies, in the support and
deployment of high-end computing systems like those currently provided
by the National Center for Supercomputing Applications (NCSA), the San
Diego Supercomputing Center (SDSC), the Pittsburgh Supercomputing
Center (PSC) and the National Center for Atmospheric Research (NCAR) to
meet the research and education user needs of the broad science and
engineering community.
Historically, mission-oriented agencies such as DOD and DOE have
driven investment in the development of operational high-end
architectures, with NSF providing the longer-term, higher-risk
investments in the basic research and education that must support these
operational developments. Industry has responded to the development of
new architectures primarily in response to markets, either commercial
or governmental. This balanced approach, with swings back and forth
between more or less investment in the long-term, has worked well in
the past. With appropriate adjustments in response to validated needs,
we believe it will continue to work well in the future.
Q2. Different scientific applications need different types of
supercomputers, and scientists from many fields use the supercomputing
capabilities supported by NSF within CISE. What role does input from
these different scientific communities play in the decisions about what
types of high-performance computing capabilities are supported by CISE?
A2. As the question indicates, different scientific applications need
and prefer different types of supercomputing capabilities. NSF's
support for high performance computing in service of science and
engineering research and education has been driven by the diverse
scientific opportunities and applications of the user community.
Informed by user community needs, NSF supports a wide range of high
performance computing system architectures.\1\ The centers and other
facilities supported by NSF are close to the end-users and thus the
decisions as to what capabilities to provide are being made as close as
possible to the community.
---------------------------------------------------------------------------
\1\ Experience has shown that the needs of the community are quite
diverse, and cannot effectively be met by providing only a single type
of system.
---------------------------------------------------------------------------
NSF continues to be the largest provider of access to a diverse set
of supercomputing platforms for open academic science and engineering
research and education in the U.S. In FY 2002, NSF provided over 152
million normalized service units to over 3,000 users involved in 1200
active awards. NSF's user community includes large numbers of NIH-,
NASA- and DOE-funded scientists and engineers.
Examples of the range of computing platforms provided through NSF
support are described below:
1. Many NSF users prefer shared memory architecture systems
with from 32 to 128 processors with large amounts of memory per
processor, since their research codes are not scalable to
several thousand processors. That is why NSF provides access to
a significant number of large memory 32-way IBM Power 4
systems, all of which are continuously over-subscribed. A new
128 way SMP HP Marvel system is also being installed at PSC.
2. At present, there are on the order of 30 research groups
that use NSF-supported supercomputers and have developed highly
scalable codes capable of efficiently using systems comprised
of thousands of processors, such as the NSF-supported Terascale
Computing System (TCS) at PSC. High allocation priority is
given to researchers who have developed such research codes. In
FY 2002, for example, 5 users accounted for 61 percent of the
usage of TCS, for projects in particle physics, materials
research, biomolecular simulations and cosmology.
3. Driven by a user community need to run very long jobs
requiring large numbers of processors, NCSA, with NSF funding,
will deploy an Intel Xeon-based Linux cluster with a peak
performance of 17.7 teraflops. This 2,900 processor Linux-based
system will be dedicated to users who need large numbers of
processors for simulations that may require up to weeks of
dedicated time.
4. Another new system, driven again by user needs, is the SDSC
DataStar, a 7.5 teraflop/s IBM Regatta system that will be
installed this fall. This system will leverage SDSC's
leadership in data and knowledge systems to address the growing
importance of large-scale data in scientific computing. The new
system will be designed to flexibly handle both data-intensive
and traditional compute-intensive applications. SDSC's Storage
Area Network, or SAN, will provide 500 terabytes of online
disk, and six petabytes of archival storage.
5. As one of the world's first demonstrations of a
distributed, heterogeneous grid computing system, the NSF-
supported Extensible Terascale Facility (ETF) will provide
access to over 20 teraflops of computing capability by the end
of FY 2004. ETF provides a distributed grid-enabled
environment, with the largest single compute cluster being a 10
teraflop IA-64 Madison cluster at NCSA. This system will
provide an integrated environment that provides unparalleled
scientific opportunity to the science and engineering
community.
Q3. What is the level of demand for cycle time on each of the NSF-
supported supercomputers? Which scientists and applications are users
of the most time? Is the demand growing? Do you have a plan to provide
the capabilities to meet existing and future levels of demand?
A3. Table 1 below describes current demand based on the number of CPU
hours available, the number requested and the number of CPU hours
allocated during FY 2003. The resources are allocated by a National
Resource Allocation Committee (NRAC) that meets twice per year. Table 1
contains allocations for all CISE-supported systems, some of which are
located at SDSC, NCSA and PSC, others of which are located at other
partner sites. In general the ratio of requested to allocated CPU hours
ranges from 1.4-2.1. Additionally, while it is not a PACI site, the
National Center for Atmospheric Research (NCAR) supercomputer has a Top
500 ranking of 13.\2\
---------------------------------------------------------------------------
\2\ http://www.top500.org/dlist/2003/06/
Table 2 overleaf provides a ranked listing of the top 25 users who
received allocations on these systems\3\ in FY 2002. Their utilization
is quoted in normalized service units, taking into account the
differential capabilities of the systems. Each investigator generally
has accounts on multiple systems. They cover a broad range of
disciplines including: particle physics, protein biomolecular
simulations, severe storm prediction, and engineering. Many of the
users of NSF-supported supercomputing centers receive research funding
from NIH and DOE. Many of the DOE-funded researchers also obtain
significant allocations through NERSC at LLBL.
---------------------------------------------------------------------------
\3\ These resources are under-subscribed by the geosciences
research community since the majority of that community uses NSF-
supported NCAR supercomputing capabilities.
Demand for supercomputing resources continues to grow each year. It
has increased from 20 million CPU hours in FY 1999 to over 60 million
CPU hours in FY 2002. During that same period of time the capacity
within NSF-supported supercomputer centers has increased from 10
million CPU hours in FY 1999 to 45 million in FY 2002. Anticipated
capacity in 2003 is 62 million CPU hours. With the installation of the
Dell cluster at NCSA in the current FY, the installation of DataStar at
SDSC by early next year, and the completion of ETF construction by the
end of FY 2004, the combined capacity will grow by an additional 55
---------------------------------------------------------------------------
million CPU hours (almost doubling the capacity) by the end of FY 2004.
Q4. Do you plan to provide funding for the purchase of new
supercomputers for the supercomputing centers?
A4. NSF's commitment to providing enhanced support for state-of-the-art
high-performance computing, in the context of cyberinfrastructure,
remains exceedingly strong. Following the trends of the last few
decades, the agency anticipates providing strong support for technology
refresh and upgrade of existing NSF-supported high performance
computing resources for the foreseeable future.
Q5. How is NSF working with the department of Energy (DOE) Office of
Science to ensure that scientists have access to a full range of high-
performance computing capabilities? What are the short-term and long-
term plans to provide this full range of capabilities?
A5. Building on a strong interagency partnership that focuses on high-
end computing, NSF is continuing to work closely with DOE to provide
high performance computing capabilities to the broad science and
engineering community. NSF and DOE staff meet regularly through both
formal (like NITRD and the High-End Computing Revitalization Task
Force) and informal processes to discuss and coordinate current and
future plans.
Demonstrating the strength of this relationship, in the first
instantiation of ETF, NSF made an award to four institutions that
included Argonne National Laboratory (ANL). The ANL component of ETF
supported the construction of a 1.25 teraflop IA-64 system with a 96-
processor visualization engine and 20 TB of storage. The award also
drew upon essential grid computing expertise available at ANL. In FY
2002 NSF funded the second phase of ETF, in an award that also drew
upon expertise at ANL. Rick Stevens from ANL is the current ETF Project
Director and Charlie Catlett from ANL is the ETF Project Manager.
Before the end of FY 2003, NSF plans to make an award to Oak Ridge
National Laboratory, to interconnect the DOE-funded Spallation Neutron
Source and other resources available on the Extensible Terascale
Facility. This award, as with all NSF awards, was identified through
merit review and will provide unique scientific opportunities for the
science and engineering user community.
NSF has already demonstrated a willingness to share NSF-supported
computational resources with a number of other federal agencies,
including DOE and NIH. For example, the 89 PIs submitting proposals for
review to the FY 2003 NRAC allocation meetings cited the following
sources of research funding (several PIs had multiple sources of
funding): NSF (55), NASA (19), NIH (19), DOE (18), DOD (11), DARPA (4),
NOAA (1), NIST (1), and EPA (1). Five of the PIs listed DOE as their
sole source of research support, three listed NIH as their sole source
of support and two listed NASA as their sole source of support.
Thirteen PIs listed no federal sources of support.
Questions submitted by Representative Ken Calvert
Q1. Cyberinfrastructure is by its nature pervasive, complex, long-
term, and multi-institutional. The National Science Foundation's
(NSF's) cyberinfrastructure effort will require persistence through
many machine life-cycles and software evolutions, and across multiple
institutions and the staff associated with them.
Q1a. How will the programs that NSF's Computer and Information Science
and Engineering (CISE) Directorate creates to accomplish its
cyberinfrastructure vision address the challenges associated with
complex programs and how will the directorate manage the multi-
institutional and long-term collaborative projects?
A1a. Dating back to the 1980's, NSF has sought to match the supply and
power of computational services to the demands of the academic
scientific and engineering community. The increasing importance of
computation and now cyberinfrastructure to scientific discovery,
learning and innovation, has guided our strategy to accomplish this.
This increasing importance, which has been fueled by advances in
computing-communications and information technologies, has led to a
number of programmatic changes over the years. For instance, the
Partnerships for Advanced Computational Infrastructure (PACI) program
was developed to meet an increased demand for computing-related
services and assistance to accompany the raw computing power
enhancements that technological innovation had provided.
The formulation of a cyberinfrastructure vision represents a new
set of opportunities, driven by the needs of the science and
engineering community and the capabilities that further technological
innovation have provided. NSF will build upon the scientific and
programmatic expertise and capability developed over the past few
decades to meet the needs of the largest number of science and
engineering researchers and educators.
Promising management approaches, designed to engage multiple
institutions in long-term collaborative projects, are currently being
discussed in collaboration with the science and engineering community.
A number of workshops and town hall meetings, which included
representatives from academe, other federal agencies and international
organizations, were held over the summer of 2003 to discuss promising
approaches.
Q1b. How will NSF and CISE address the need to provide support
overtime and institutions so that investments in facilities and
expertise can effectively be leveraged?
A1b. As indicated above, promising management approaches, designed to
engage multiple institutions in long-term collaborative projects, are
currently being discussed in collaboration with the science and
engineering community. A number of workshops and town hall meetings,
which included representatives from academe, other federal agencies and
international organizations, were held over the summer of 2003 to
discuss promising approaches. Our goal is to use the most promising
approaches identified to ensure that investments in facilities and
expertise can be effectively leveraged.
Q2. NSF has already announced that the current Partnership for
Advanced Computational Infrastructure (PACI) program will end at the
end of fiscal year 2004, and that the new program for TeraGrid partners
will begin in fiscal year 2005. However this is less than 15 months
away and the supercomputing centers have not heard information about
the objectives, structure or recommended budgets for the program.
Q2a. LWhat is the schedule for developing these plans and what can you
tell us now about plans for this or other related programs?
A2a. NSF plans to issue detailed guidance to SDSC, NCSA and PSC during
the fall of 2003. Guidance will include discussion of means of support
to be provided for both ETF, in which SDSC, PSC and NCSA are partners
together with other organizations, and for other high performance
computing resources and services provided by these centers under NSF
support. Some discussions with senior SDSC, PSC and NCSA personnel have
already taken place.
Q2b. Will NSF's plans be consistent with the recommendation from the
Atkins report, which states in Section 5.3 that ``the two existing
leading-edge sites (the National Center for Supercomputing Applications
and the San Diego Supercomputing Center) and the Pittsburgh
Supercomputing Center should continue to be assured of stable,
protected funding to provide the highest-end computing resources?
A2b. NSF plans are consistent with the recommendations of the Atkins
report in that stable funding for NCSA, SDSC and PSC is anticipated to
provide high performance computing capabilities to the science and
engineering community.
Q2c. The Atkins report also assumes the centers would operate with an
annual budget of approximately $75 million per center. Is this the
direction NSF envisions for this program and if so, when is this level
of support expected to start and how long will it last?
A2c. NSF is currently developing a five-year planning document for
cyberinfrastructure. The development of this plan is informed by the
recommendations of the Atkins report, as well as other community input.
The first edition of the plan, which will be a living document, should
be available in 2004.
Questions submitted by Representative Ralph M. Hall
Q1. The National Science Foundation (NSF) blue ribbon panel on
cyberinfrastructure needs, the Atkins Panel, which issued its report
this past February, emphasized that NSF has an important role in
fostering development and use of high performance computers for broad
research and education in the sciences and engineering. The Atkins
report strongly recommended that U.S. academic researchers should have
access to the most powerful computers at any point in time, rather than
10 times less powerful, ``as has often been the case in the last
decade,'' according to the report. To provide the research community
with this level of capability, the report recommends funding 5 centers
at a level of $75 million each, of which $50 million would be for
hardware upgrades needed to acquire a major new system every 3 or 4
years.
Q1a. What is your response to this recommendation that NSF provide the
resources necessary to ensure that high-end computing systems are
upgraded with regularity to insure that the research community has
access to leading edge computers at all times?
A1a. NSF recognizes that technology refresh is very important in order
to keep high-end computing resources and hence related science and
engineering research and education, at the state of the art. Over the
past five years, NSF has demonstrated its commitment to doing so. Table
3 below provides evidence that this is the case.
Q1b. What level of funding has NSF provided over the past 5 years for
upgrading the high-end computer systems at its major computer centers?
What is NSF's current plan for support of high-end computer centers for
providing hardware upgrades needed to keep them at the leading edge
(break out funding by categories for operations and maintenance and
hardware upgrades)?
A1b. With funding provided through the PACI program and through the
Terascale Initiative, NSF has invested over $210 million on hardware
and integrated software upgrades and renewal over the past five years.
The agency remains committed to technology refresh and upgrades,
recognizing that this is essential to realize the promise of the
cyberinfrastructure vision.
Q1c. You have announced a reorganization of the Computer and
Information Science and Engineering Directorate that combines the
computer and network infrastructure divisions. Will support for the
operation of state-of-the-art high-end computing facilities continue
under this reorganization, or does this signal a change in the
priorities? Will funding for FY 2003 and FY 2004 for Partnerships for
Advanced Computational Infrastructure (PACT) centers increase, decrease
or stay the same under this reorganization relative to FY 2002 funding
levels?
A1c. The reorganization of CISE does not signal a change in priorities.
It has been proposed in order to support the ever-broadening meaning of
``state-of-the-art high-end computing facilities'' that the
cyberinfrastructure vision illuminates. Support for operations of NSF's
state-of-the art high-end computing facilities will continue and grow
under this reorganization. The reorganization will provide the CISE
directorate with a more focused and integrated organization to deal
with the deployed computational, networking, storage and middleware
infrastructure essential to cyberinfrastructure.
Q2. You pointed out in your testimony that high-end computing is only
one component of deploying an advanced cyberinfrastructure needed for
the advancement of science and engineering research and education. Are
you satisfied with the priority afforded high-end computing in the
current NSF budget that supports cyberinfrastructure development and
deployment?
A2. High-end computing is essential to the progress of science and
engineering. NSF's budget requests and investments are designed to
recognize the crucial role cyberinfrastructure plays in enabling
discovery, learning, and innovation across the science and engineering
frontier. While need continues to outstrip the available resources, NSF
budget requests continue to be very responsive to the needs of the
community.
Q3. What do you see NSF's role relative to the Defense Advanced
Research Projects Agency and the Department of Energy in supporting
research related to the development and use of future U.S. academic
research community? What portion of your directorate's budget would you
expect to allocate for these purposes?
A3. As in other fields of science and engineering, NSF draws upon
partnership efforts to leverage its investments in cyberinfrastructure.
For example, the agency's partnership with other federal agencies in
the area of high-end computing and networking is nurtured through the
High End Computing program component area of the Networking and
Information Technology Research and Development (NITRD) Working Group.
This partnership has recently been strengthened through the focused
activities of the High-End Computing Revitalization Task Force.
These partnership activities strengthen the management and
coordination of relevant programs and activities to increase the return
on current investments and to maximize the potential of proposed
investments. NSF leverages nearer-term high-end computing research and
development programs funded by DARPA, DOD and DOE, where government-
industry partnerships often create new generations of high-end
programming environments, software tools, architectures, and hardware
components to realize high-end computing systems that address issues of
low efficiency, scalability, software tools and environments, and
growing physical constraints. By drawing upon its effective interagency
relationships, NSF avoids duplication of effort. NSF focuses its
investments in higher-risk, longer-term research investments to ensure
that the new innovation is possible years from now. Examples of current
high-risk, longer-term basic research showing promise for high
performance computing applications include nanoscale science and
engineering and bio-inspired computing.
Finally, NSF capitalizes upon the outcomes of Federal Government-
supported nearer-term high-end computing research and development
activities enabled by its sister agencies, in the support and
deployment of high-end computing systems like those currently provided
by the National Center for Supercomputing Applications (NCSA), the San
Diego Supercomputing Center (SDSC), the Pittsburgh Supercomputing
Center (PSC) and the National Center for Atmospheric Research (NCAR) to
meet the research and education user needs of the broad science and
engineering community.
In preparing its annual budget request, the agency gives careful
consideration to funding provided for cyberinfrastructure. As the
agency focuses increasing attention on cyberinfrastructure, it is
likely that the funds dedicated to the development of an enabling,
coherent, coordinated cyberinfrastructure portfolio will grow, in
recognition of the importance of cyberinfrastructure to all of science
and engineering.
Answers to Post-Hearing Questions
Responses by Daniel A. Reed, Director, National Center for
Supercomputing Applications, University of Illinois at Urbana-
Champaign
Questions submitted by Representative Ralph M. Hall
Q1. You cited in your testimony the National Science Foundation (NSF)
blue ribbon panel report that recommended a funding level of $75
million per year to enable a supercomputer center to always have a
state-of-the-art computer. What is the annual funding level provided by
NSF for your center at present, and how has it varied over time? What
has been the funding trend for hardware upgrades?
A1. For FY 2003, we anticipate receiving $34,650,00 for the sixth year
of the NSF cooperative agreement for the National Computational Science
Alliance (Alliance), one of two Partnerships for Advanced Computational
Infrastructure (PACI).\1\ I have attached a chart that shows the
funding history for NCSA (the leading edge site of the Alliance) and
the Alliance from FY 1998 through FY 2004 (which is an estimate based
on our program plan for the coming year).
---------------------------------------------------------------------------
\1\ The cooperative agreement for the other partnership, the
National Partnership for Advanced Computational Infrastructure (NPACI)
is similar.
---------------------------------------------------------------------------
As the chart illustrates, NSF funding for the Alliance was $29
million in FY 1998, $34.3 million in FY 1999, $33.7 million in FY 2000,
$35.17 million in FY 2001 and $35.25 million in FY 2002. The annual
funding level has remained relatively flat throughout the program's
lifetime, and we anticipate the FY 2004 total will be $33.5 million,
about $1 million less than FY 2003.
I have also included the budget for hardware during this six-year
period. In the initial years of the program, software and maintenance
were included in the hardware budget line; this cost was moved to the
operations budget in FY 2000. As one can see, the hardware budget
includes not only funds for supercomputing (capability computing)
hardware, but also support for data storage, networking, desktop
support, Access Grid (distributed collaboration) and audio/visual
support, and testbed systems.
The support specifically for supercomputing hardware has varied. In
FY 2001, supercomputing hardware was funded at $9.08 million, in the
current year $11.795 million has been allocated, and the FY 2004
request is for $8.63 million. Funding for hardware upgrades (i.e., the
annual hardware budget) has remained fat, varying between $13 and $15
million.
It is important to note that when the PACI program began, its
participants expected overall NSF support to increase annually during
the program's ten-year lifetime (i.e., at a minimum to reflect standard
inflation). The original request for FY 1998 from NSF was $34.988
million, and we anticipated it would grow to $49.271 million by FY
2002. That steady growth never materialized, despite substantial
increases in human resource costs during that time. Hence, the Alliance
and NCSA (and its sister institutions SDSC and NPACI) have experienced
de facto annual budget cuts.
In addition, NCSA, in collaboration with partners at the San Diego
Supercomputing Center, Argonne National Laboratory, and the California
Institute of Technology, was chosen by NSF to create the Distributed
Terascale Facility or TeraGrid.\2\ This three-year award, totaling $53
million, has provided $19,450,500 to NCSA for hardware, storage,
networking and software support. This is, however, substantially less
than an annual ``cost of living'' adjustment would have provided in
aggregate.
---------------------------------------------------------------------------
\2\ The Pittsburgh Supercomputing Center was added in 2002.
Q2. A reorganization has been announced for the NSF Computer and
Information and Engineering Directorate that combines the computer and
network infrastructure divisions. Have you seen any evidence of how
this reorganization will affect the Partnerships for Advanced
Computational Infrastructure (PACI) program, and your center in
---------------------------------------------------------------------------
particular?
A2. The NSF's announcement of the CISE Directorate reorganization is
relatively recent. There are not yet enough organizational or funding
details for substantive assessment.
NSF has announced that the PACI program will end on September 30,
2004--just over one year from now. We do not know, with any
specificity, the nature of NSF's plans for high-end computing in FY
2005 and beyond. The cyberinfrastructure panel, which was charged by
NSF with evaluating the centers and outlining future directions,
recommended long-term, stable funding for the existing centers and an
expansion of investments in computing infrastructure. However, to date,
we have no specific information on NSF plans, out-year budget requests,
or the process by which funding would be secured.
With the announced end of the PACI program only a year away and the
details of successor structures not yet known, NCSA, our partner
institutions within the Alliance, and those at the San Diego
Supercomputer Center and the Pittsburgh Supercomputing Center, remain
concerned about the future. We strongly believe NSF must continue to
make significant annual investments in high-end computing if the
national research community is to have access to the most powerful
computational resources. Only with these investments can the research
community address breakthrough scientific problems and maintain
international leadership.
Q3. You pointed out the importance of deep involvement and coordinated
collaboration of computer vendors, national labs and center, and
academic researchers, along with multi-agency investments, to develop
and deploy the next generation of high-end computer systems. What is
your assessment of the current state of coordination in this area, and
what recommendations do you have on how to improve the situation?
A3. During the early days of the HPCC program, cross-agency activities
were somewhat more coordinated than they are now. Hence, my personal
view is that there has not been coordinated collaboration on the
development of high-end computing systems across the Federal Government
for several years. Some of this compartmentalization is understandable,
given the differing missions and goals of the several agencies
(Department of Defense, National Security Agency, Department of Energy,
National Science Foundation, National Aeronautics and Space
Administration, the National Institutes of Health and others) that
sponsor research in high-end computing and acquire and utilize high-end
computing systems.
As I testified at the hearing, the activities of the current High
End Computing Revitalization Task Force (HECRTF) are encouraging.
Charting a five-year plan for high-end computing across all relevant
federal agencies involved in high-end computing research, development,
and applications is critical to the future of high-end computing
research, computational science and national security. I was pleased to
lead the community input workshop for HECRTF, and I am hopeful the
views and suggestions of research community will be incorporated into
the vision and recommendations of the Task Force.
However, I believe success rests on more than budgetary
coordination. We must also ensure that insights and promising ideas
from basic research in high-performance computing are embraced and
developed as advanced prototypes. Successful prototypes should then
transition to production and procurement. Greater interagency
coordination is needed to ensure such transitions occur.
Appendix 2:
----------
Additional Material for the Record
Additional Statement by Dr. Raymond L. Orbach
There are two aspects to this question: the extent to which
supercomputing will aid or accelerate development of nanotechnology
(through, e.g., modeling and simulation of nanostructures), and the
extent to which nanotechnological advances will contribute to
supercomputing in Japan.
With respect to the first issue, there is no question that
supercomputers will allow more detailed, complete, and accurate
modeling of larger collections of atoms, and permit simulations to
cover longer time periods. Both of these capabilities are critical in
connecting basic theory to practical experimental data about assembly,
structure, and behavior of materials at the nanoscale. There are in
fact research projects using Japan's Earth Simulator that address such
questions, such as one entitled ``Large-scale simulation on the
properties of carbon-nanotube'' (Kazuo Minami, RIST).
With respect to the impact of Japanese nanotechnology on Japanese
supercomputing, the technology appears to not be sufficiently mature
for a major impact in the next few years. Nanotechnology products are
beginning to head into the marketplace and will affect the computing
industry; examples can certainly be found both in Japanese and in
American companies. (From Japan, NEC recently announced that they
intend to use fuel cells based on a form of carbon nanotube to extend
battery cycles in notebook computers from 4 to 40 hours. A recent U.S.
example is the development by Motorola of a technology to produce large
flat panel displays based on electron emission from carbon nanotubes,
for which they are in the process of negotiating licensing agreements.)
However, these are currently targeted at the consumer electronics
market and may not have immediate impact on supercomputing. For the
latter, developments in areas such as heat conduction using nanoscale
technologies may have an impact by facilitating cooling of
supercomputer architectures.
Beyond the end of the silicon semiconductor ``roadmap,'' in another
10-15 years, the development of molecular electronics may begin to have
an impact on all forms of computing. Predictive computer modeling of
molecular electronics may well be essential for design and
manufacturability of such structures, just as computer-aided design has
proven critical to the development of current-day circuits.