[House Hearing, 114 Congress]
[From the U.S. Government Publishing Office]
SUPERCOMPUTING AND AMERICAN
TECHNOLOGY LEADERSHIP
=======================================================================
HEARING
BEFORE THE
SUBCOMMITTEE ON ENERGY
COMMITTEE ON SCIENCE, SPACE, AND TECHNOLOGY
HOUSE OF REPRESENTATIVES
ONE HUNDRED FOURTEENTH CONGRESS
FIRST SESSION
__________
JANUARY 28, 2015
__________
Serial No. 114-03
__________
Printed for the use of the Committee on Science, Space, and Technology
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Available via the World Wide Web: http://science.house.gov
______
U.S. GOVERNMENT PUBLISHING OFFICE
93-881PDF WASHINGTON : 2015
-----------------------------------------------------------------------
For sale by the Superintendent of Documents, U.S. Government Publishing
Office Internet: bookstore.gpo.gov Phone: toll free (866) 512-1800;
DC area (202) 512-1800 Fax: (202) 512-2104 Mail: Stop IDCC,
Washington, DC 20402-0001
COMMITTEE ON SCIENCE, SPACE, AND TECHNOLOGY
HON. LAMAR S. SMITH, Texas, Chair
FRANK D. LUCAS, Oklahoma EDDIE BERNICE JOHNSON, Texas
F. JAMES SENSENBRENNER, JR. ZOE LOFGREN, California
DANA ROHRABACHER, California DANIEL LIPINSKI, Illinois
RANDY NEUGEBAUER, Texas DONNA F. EDWARDS, Maryland
MICHAEL T. McCAUL FREDERICA S. WILSON, Florida
STEVEN M. PALAZZO, Mississippi SUZANNE BONAMICI, Oregon
MO BROOKS, Alabama ERIC SWALWELL, California
RANDY HULTGREN, Illinois ALAN GRAYSON, Florida
BILL POSEY, Florida AMI BERA, California
THOMAS MASSIE, Kentucky ELIZABETH H. ESTY, Connecticut
JIM BRIDENSTINE, Oklahoma MARC A. VEASEY, Texas
RANDY K. WEBER, Texas KATHERINE M. CLARK, Massachusetts
BILL JOHNSON, Ohio DON S. BEYER, JR., Virginia
JOHN R. MOOLENAAR, Michigan ED PERLMUTTER, Colorado
STEVE KNIGHT, California PAUL TONKO, New York
BRIAN BABIN, Texas MARK TAKANO, California
BRUCE WESTERMAN, Arkansas BILL FOSTER, Illinois
BARBARA COMSTOCK, Virginia
DAN NEWHOUSE, Washington
GARY PALMER, Alabama
BARRY LOUDERMILK, Georgia
------
Subcommittee on Energy
HON. RANDY K. WEBER, Texas , Chair
DANA ROHRABACHER, California ALAN GRAYSON, Florida
RANDY NEUGEBAUER, Texas DANIEL LIPINSKI, Illinois
MO BROOKS, Alabama ERIC SWALWELL, California
RANDY HULTGREN, Illinois ELIZABETH H. ESTY, Connecticut
THOMAS MASSIE, Kentucky MARC A. VEASEY, Texas
BARBARA COMSTOCK, Virginia KATHERINE M. CLARK, Massachusetts
DAN NEWHOUSE, Washington EDDIE BERNICE JOHNSON, Texas
BARRY LOUDERMILK, Georgia
LAMAR S. SMITH, Texas
C O N T E N T S
January 28, 2015
Page
Witness List..................................................... 2
Hearing Charter.................................................. 3
Opening Statements
Statement by Representative Randy K. Weber, Chairwoman,
Subcommittee on Energy, Committee on Science, Space, and
Technology, U.S. House of Representatives...................... 5
Written Statement............................................ 6
Statement by Representative Eddie Bernice Johnson, Ranking
Member, Committee on Science, Space, and Technology, U.S. House
of Representatives............................................. 6
Written Statement............................................ 7
Witnesses:
Mr. Norman Augustine, Board Member, Bipartisan Policy Center
Oral Statement............................................... 9
Written Statement............................................ 12
Dr. Roscoe Giles, Chairman, DOE Advanced Scientific Computing
Advisory Committee
Oral Statement............................................... 17
Written Statement............................................ 19
Mr. David Turek, Vice President, Technical Computing, IBM
Oral Statement............................................... 50
Written Statement............................................ 52
Dr. James Crowley, Executive Director, Society for Industrial and
Applied Mathematics
Oral Statement............................................... 59
Written Statement............................................ 61
Discussion....................................................... 66
Appendix I: Answers to Post-Hearing Questions
Mr. Norman Augustine, Board Member, Bipartisan Policy Center..... 76
Dr. Roscoe Giles, Chairman, DOE Advanced Scientific Computing
Advisory Committee............................................. 77
Mr. David Turek, Vice President, Technical Computing, IBM........ 82
Dr. James Crowley, Executive Director, Society for Industrial and
Applied Mathematics............................................ 86
SUPERCOMPUTING AND AMERICAN TECHNOLOGY LEADERSHIP
----------
WEDNESDAY, JANUARY 28, 2015
House of Representatives,
Subcommittee on Energy
Committee on Science, Space, and Technology,
Washington, D.C.
The Subcommittee met, pursuant to call, at 9:08 a.m., in
Room 2318 of the Rayburn House Office Building, Hon. Randy
Weber [Chairman of the Subcommittee] presiding.
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman Weber. Well, good morning and welcome to today's
Energy Subcommittee hearing titled ``Supercomputing and
American Technology Leadership.''
The Committee will come to order.
Without objection, the Chair is authorized to declare
recesses of the Subcommittee at any time.
Without objection, the Chair authorizes the participation
of Mr. Lipinski, Mr. Swalwell, Mr. Grayson, Ms. Esty, Mr.
Veasey, and Ms. Clark for today's hearing. And I understand
Ranking Member Johnson will serve as the Ranking Minority
Member today and give an opening statement a little later.
In front of you are packets containing the written
testimonies, biographies, and truth-in-testimony disclosures
for today's witnesses. And I recognize myself for five minutes
for an opening statement.
At the outset let me say that this is my first Committee
hearing as a Chairman of this Subcommittee and it is truly an
honor to be selected to serve in this capacity. And I want to
say a personal thanks to Chairman Lamar Smith for his help and
his guidance. He has been just a stalwart friend of mine. I
really appreciate that.
This Committee will tackle a number of important issues
related to America's competitiveness and energy future, and I
am excited to be part of these important discussions.
Today, we are going to hear from a distinguished panel of
witnesses about the importance of high-performance computing to
American technological competitiveness, specifically focusing
on the Department of Energy's Advanced Scientific Computing
Research program, also known as the ASCR program within the
Office of Science.
High-performance computing provides a platform for
breakthroughs in all scientific research and accelerates
applications of scientific breakthroughs across our economy.
Progress in computing has paved the way for breakthroughs in
medical imaging, genetics research, manufacturing, engineering,
and weapons development. Faster computing speeds have
revolutionized the energy sector, improving the efficiency of
energy production and aiding in distribution technologies.
Advances in modeling and algorithm development offer
opportunities for scientific discovery in fields where
experiments are too difficult, too costly, or too dangerous to
conduct. They are reducing costs and opening the door to more
innovative discoveries.
The work underway in the ASCR program drives breakthroughs
in high-performance computing. The Department of Energy's
national labs host world-class computational science
facilities, and the Department funds the applied mathematical
and computational science research that will drive the next
stage of advancement in this field.
As we face the reality of ongoing budget constraints in
Washington, it is our job in Congress to ensure that taxpayer
dollars are spent wisely on innovative research that is in the
best national interest and provides the best chance for broad
impact and long-term success. The basic research conducted
within the ASCR program clearly meets this requirement. High-
performance computing can lead to scientific discoveries,
economic growth, and will help maintain America's leadership in
science and technology.
I want to thank the witnesses in advance for participating
in today's hearing and look forward to further discussion.
[The prepared statement of Mr. Weber follows:]
Prepared Statement of Subommittee Chairman Randy Weber
Good morning and welcome to today's Energy Subcommittee hearing
titled ``Supercomputing and American Technology Leadership.''
Today, we will hear from a distinguished panel of witnesses about
the importance of high performance computing to American technological
competitiveness, specifically focusing on the Department of Energy's
Advanced Scientific Computing Research program, also known as the
``ASCR'' program within the Office of Science.
High performance computing provides a platform for breakthroughs in
all scientific research, and accelerates applications of scientific
breakthroughs across our economy. Progress in computing has paved the
way for breakthroughs in medical imaging, genetics research,
manufacturing, engineering, and weapons development. Faster computing
speeds have revolutionized the energy sector, improving the efficiency
of energy production and aiding in distribution technologies. Advances
in modeling and algorithm development offer opportunities for
scientific discovery in fields where experiments are too difficult,
costly, or dangerous to conduct, reducing costs and opening the door to
more innovative discoveries.
The work underway in the ASCR program drives breakthroughs in high
performance computing. The Department of Energy's national labs host
world-class computational science facilities, and the department funds
the applied mathematical and computational science research that will
drive the next stage of advancement in this field.
As we face the reality of ongoing budget constraints in Washington,
it is our job in Congress to ensure that taxpayer dollars are spent
wisely, on innovative research that is in the national interest, and
provides the best chance for broad impact and long-term success. The
basic research conducted within the ASCR program clearly meets this
requirement. High performance computing can lead to scientific
discoveries, economic growth, and will maintain America's leadership in
science and technology. I thank the witnesses for participating in
today's hearing and look forward to further discussion.
Chairman Weber. I now recognize Ranking Member Johnson for
an opening statement.
Ms. Johnson. Thank you very much, Mr. Chairman, and I thank
you for holding this hearing. And I want to thank our very
excellent panel of witnesses for their testimony and being here
today.
America has historically been a leader in advancing new
energy technologies, as well as the fundamental sciences of
physics, chemistry, engineering, mathematics, and computational
science that support energy innovation. But our leadership in
technology is challenged by the growing investments of other
countries in education and research, investments that are now
predicted to quickly outpace our own investments here at home.
High-performance computing or supercomputing is one area
that we have led in for decades and the United States currently
holds more than 45 percent of the 500 fastest supercomputers in
the world. These computers are capable of processing vast
amounts of data and mathematical equations at amazing speeds.
In the past, high-performance computers were needed
primarily for specialized scientific and engineering
applications. Now, as we enter the world of big data where
thousands of devices all around us are generating millions of
bytes of data to be analyzed, high-performance computing is
needed not just by scientists and government researchers but by
many civic and commercial enterprises as well.
Public policies play a critical role in supporting the
advancement of high-performance computing and in enabling our
society and economy to directly benefit from this capability.
Our policies allow researchers and private industry to access
the Department of Energy's computing systems, which are some of
the most powerful in the world. We set policies that support
the development of the software necessary to operate and
optimize the use of high-performance systems, software that is
unlikely to be developed by private industry because the
potential sales market is too small to support the initial
research and development costs. And our policies ensure that
our investments in new computer architectures are diverse and
flexible enough to meet our national security needs, in
addition to our research and private industry needs. Federal
investments in high-performance computing open this technology
up to the future development of proprietary products. They grow
our technology economy and they advance our technological
leadership internationally.
Now, while every witness on this panel is extremely
distinguished and I am grateful that each of you could be here
today, I hope you won't mind if I thank Dr. Augustine in
particular for taking time to speak with us this morning as he
has been a great friend to this Committee for well over a
decade. As a former Chairman of Lockheed Martin and the Chair
of the National Academy of Sciences Committee that produced the
seminal Rising above the Gathering Storm report in 2005, he has
a broad and deep perspective on the challenges facing our
Nation in research and technological innovation. That report
laid the foundation for one of our Committee's landmark
bipartisan achievements, the America COMPETES Act of 2007,
which we reauthorized in 2010 and I hope the next
reauthorization is a top priority for the Committee and this
Congress.
I look forward to hearing Mr. Augustine's thoughts and
indeed those of all of our witnesses on where we need to go in
scientific research and innovation to grow our economy and to
improve the quality of life for all Americans. Working
together, our Committee has the opportunity to renew our
commitment to scientific and technological leadership by our
actions, and I look forward to any input our panelists have
toward that goal.
With that, I thank you for coming and I yield back the
balance of my time.
[The prepared statement of Ms. Johnson follows:]
Prepared Statement of Committee on Science, Space, and Technology
Ranking Member Eddie Bernice Johnson
Thank you Chairman Weber for holding this hearing, and I
also want to thank this excellent panel of witnesses for their
testimony and for being here today.
America has historically been a leader in advancing new
energy technologies, as well as the foundational sciences of
physics, chemistry, engineering, mathematics, and computational
science that support energy innovation. But our leadership in
technology is challenged by the growing investments of other
countries in education and research; investments that are now
projected to quickly outpace our own investments here at home.
High performance computing, or supercomputing, is one area
we have led in for decades, and the U.S. currently hosts more
than 45% of the 500 fastest supercomputers in the world. These
computers are capable of processing vast amounts of data and
mathematical equations at amazing speeds. In the past, high
performance computers were needed primarily for specialized
scientific and engineering applications. Now, as we enter the
world of `big data', where thousands of devices all around us
are generating millions of bytes of data to be analyzed, high
performance computing is needed not just by scientists and
government researchers, but by many civic and commercial
enterprises as well.
Public policies play a critical role in supporting the
advancement of high performance computing, and in enabling our
society and economy to directly benefit from this capability.
Our policies allow researchers and private industry to access
the Department of Energy's computing systems, which are some of
the most powerful in the world. We set policies that support
the development of the software necessary to operate and
optimize the use of high performance systems--software that is
unlikely to be developed by private industry because the
potential sales market is too small to support the initial
research and development costs. And our policies ensure that
our investments in new computer architectures are diverse and
flexible enough to meet our national security needs, in
addition to our research and private industry needs. Federal
investments in high performance computing open this technology
up for future development of proprietary products, they grow
our technology economy, and they advance our technological
leadership internationally.
Now, while every witness on this panel is extremely
distinguished and I am grateful that each of you could be here
today, I hope you won't mind if I thank Dr. Augustine in
particular for taking time to speak with us this morning, as he
has been a great friend to the Committee for well over a
decade. As the former Chairman of Lockheed Martin and the Chair
of the National Academy of Sciences Committee that produced the
seminal Rising Above the Gathering Storm report in 2005, he has
a broad and deep perspective on the challenges facing our
nation in research and technological innovation. That report
laid the foundation for one of our Committee's landmark
bipartisan achievements, the America COMPETES Act of 2007,
which we reauthorized in 2010, and I hope the next
reauthorization is a top priority for the Committee in this new
Congress.
I look forward to hearing Mr. Augustine's thoughts--and
indeed those of all of our witnesses - on where we need to go
in scientific research and innovation to grow our economy and
to improve the quality of life of all Americans. Working
together, our Committee has the opportunity to renew our
commitment to scientific and technological leadership by our
actions, and I look forward to any input our panelists have
towards that goal.
With that, I thank you all for coming, and I yield back the
balance of my time.
Chairman Weber. I thank the lady, and if there are Members
who wish to submit additional opening statements, your
statements will be added to the record at this point.
Chairman Weber. At this time I would like to introduce our
witnesses. Our first witness, who comes with high
commendations, is Mr. Norman Augustine, Board Member of the
Bipartisan Policy Center. Mr. Augustine served as the
Undersecretary of the Army and later as acting Secretary of the
Army from 1975 to 1977. Mr. Augustine also served as the
President and CEO of Lockheed Martin until he retired in 1997.
He has been a member of advisory boards to the Department of
Homeland Security, Energy, Defense, Commerce, Transportation,
and Health and Human Services, as well as NASA, Congress, and
the White House.
Is there any other--are there boards that you weren't a
member of, Mr. Augustine?
Our second witness today who is actually joining us by
video is Dr. Roscoe Giles, Chairman of the Advanced Scientific
Computing Advisory Committee at the Department Of Energy and a
Professor at Boston University. Dr. Giles has served in a
number of leadership roles in the community, including Member
of the Board of Associated Universities Incorporated, Chair of
the Boston University Faculty Council, and General Chair of the
SC conference in 2002. Welcome, Dr. Giles.
Dr. Giles. Thank you.
Chairman Weber. Our next witness today is Mr. David Turek,
Vice President of Technical Computing at IBM. Previously Mr.
Turek--am I saying that name correctly? Okay. Previously, Mr.
Turek helped launch IBM's grid computing business and ran IBM's
Linux cluster business. He also helped lead IBM's initiative in
support of the U.S. Accelerated Strategic Computing Initiative
at Lawrence Livermore National Laboratory, which I believe is
in Mr. Swalwell's district.
Mr. Swalwell. That is right.
Chairman Weber. Yes. So welcome.
Our final witness today is Dr. James Crowley, Executive
Director at the Society for Industrial and Applied Mathematics.
Dr. Crowley has held this position since 1995. Prior to this,
he served in the Air Force for 22 years retiring as Lieutenant
Colonel. Dr. Crowley is a fellow of the American Mathematical
Society and a fellow of the American Association for the
Advancement of Science.
In order to allow time for discussion, please limit your
testimony to five minutes, we ask the witnesses, and your
entire statement will be made part of the written record.
I now recognize Mr. Augustine for five minutes to present
his testimony.
TESTIMONY OF MR. NORMAN AUGUSTINE,
BOARD MEMBER, BIPARTISAN POLICY CENTER
Mr. Augustine. Well, thank you very much, Chairman Weber,
Ranking Member Johnson, and Members of the Subcommittee, and
thank you, Ranking Member Johnson, for all those kind words.
I am particularly appreciative that this Committee is going
to devote some time to the topic at hand and certainly high-
performance computing is a key element of research.
I will submit a statement for the record.
I would like to begin by offering a few words about the
basic nature of research. It is through research that new
knowledge is created that permits engineers like myself to
translate that research, knowledge into products and services
that, working with entrepreneurs, can go into the marketplace
and improve people's lives. We often think of Apple, the great
things it has done, deservedly. Think of the iPod, iPads, and
so on. But it wasn't Apple that made those things possible; it
was researchers working decades ago on such things as quantum
mechanics and material sciences, solid-state physics, and so
on.
One of the things about basic research in particular is
that you can't know or priority what will be the outcome of it
and that sure makes it particularly difficult in your roles, to
build support for it, yet there are so many examples of where
basic research that was curiosity-driven led to greater
improvements in people's lives. Three things that come to my
mind, one is research on seals in Antarctica that led to a
surgical procedure that saved the lives of many children
undergoing lung surgery. Another was study of the chemistry of
butterfly wings of that led to an ingredient that is used in
chemotherapy. Still another of course would be the accidental
discovery of penicillin when someone was studying research on
bacteria many, many decades ago, Sir Alexander Fleming.
I would like to quickly touch on the importance of research
and I will cite three areas where I think it has particularly
had an impact. One is on the creation of jobs and there is
evidence that if you want to one percentage point to the
average number of jobs in America, you have to add about 1.7
percentage points to the GDP of America. There have been a
number of studies, one of which was the basis of a Nobel Prize
and it has shown that between 50 and 85 percent of the growth
of GDP in our country during the last half-century is directly
attributable to advancements in two fields: science and
technology. And of course those advancements are entirely
dependent upon research.
Health is an example. In the last century life expectancy
in the United States grew from 47 to 79 years. In fact, I am 79
years old so this is really important to me. The life
expectancy gain that came about was in considerable part
attributable to advancements in biomedical research.
A third example is things that we take for granted in our
everyday life, be they television, electric cars, DVDs, GPSs,
CAT scans, or what have you, are dependent upon the knowledge
that came through basic research.
Touching briefly on high-performance computing, it impacts
field across the entire technological spectrum. My own field of
aerodynamics is an example, another would be genomics, high-
energy physics. It truly is of broad importance.
The Department of Energy, as you know, operates 17
laboratories. They are able to do things that industry really
can't do under the pressures of today's marketplace for quick
returns, financial returns. The examples, things that they
could do so well are high-risk, high-return payoff research or
long-term research, large research projects. They are
particularly well suited to that. And work in the past, for
example, sponsored by the Department of Energy on hydraulic
fracturing, as you know, has had an enormous impact today in
the political world, as well as the economic world.
How are we doing in the United States in research? The
answer has to be not very well. Research funding as a
percentage of GDP of the United States has dropped from 1st
place to 7th place in the last decade or so. The fraction of
research in a country that is sponsored by the government,
United States is down in 29th place. As a fraction of GDP--R&D
to GDP we are in 10th place now. In five years China is very
likely to pass us in research in the absolute sense and as a
fraction of GDP.
Finally, I would note that H.R. 5120 that was introduced
last year contributes in a major way to solving what I think
are some of the problems we have at translating the research
that goes on in the DOE laboratories to the commercial sector,
and I would be happy to address that further should the
Committee wish. Thank you very much.
[The prepared statement of Mr. Augustine follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman Weber. Thank you, Mr. Augustine.
And now, we recognize Dr. Giles.
TESTIMONY OF DR. ROSCOE GILES, CHAIRMAN,
DOE ADVANCED SCIENTIFIC COMPUTING
ADVISORY COMMITTEE
Dr. Giles. Thank you, Chairman Weber, and Ranking Member
Johnson, and Members of the Committee. Thank you for inviting
me to testify today and thanks for your support of the
outstanding scientific and technical activities we are here to
discuss.
The Advanced Scientific Computing Advisory Committee,
ASCAC, which I chair, is a panel of experts that advises DOE
under FACA rules about activities of the Office of Advanced
Scientific Computing Research, ASCR. My testimony is largely
based on ASCAC reports. I will address the value of research
supported directly and indirectly by ASCR and also the
technological challenges and rewards represented by U.S.
leadership in this field.
The computing needs of science have grown exponentially,
paralleling the exponential increases in computer power we have
seen in recent decades sometimes pushing the computer industry
for new capabilities and sometimes finding novel ways to
exploit existing technology. The combination of computing power
and the ability to transport, store, and learn from vast
amounts of data is critical to U.S. leadership in a wide range
of scientific and technical fields.
ASCR has enabled DOE scientists to harness unprecedented
computing power to better understand the physical world, design
new materials and devices, and engineer new and improved
methods for energy production, utilization, and distribution.
Recent examples include microscopic modeling of nuclear reactor
core startup that can improve reactor efficiency and safety;
simulations of complex combustion making the chemistry and
physics of fluids and gases to the observed behavior of engines
and reactor; predictive modeling of materials for lithium air
batteries systems potentially able to store 10 times as much
energy as lithium ion batteries; wheat genome sequencing
previously impossible to do is now possible in under 32 seconds
using new programming methods developed by ASCR; and modeling
the surface of human skin to understand its properties and how
chemicals might affect it. My written testimony includes many
additional examples.
ASCR enables such outcomes by designing and deploying an
effective system of world-class facilities for computing, data
science, and networking in DOE labs making available expert
staff to work with scientists to push the envelope of
applications and supporting research in computer science in
applied mathematics leading to key advances in software,
hardware, algorithms, and applications.
Success also depends on a knowledgeable workforce and an
educational pipeline to create that workforce. ASCR supports
both training programs for scientists and the renowned
Computational Science Graduate Fellowship program, CSGF. ASCR
nurtures all elements of the ecosystem for scientific
computing.
What about the future? ASCR has consistently provided
leadership to DOE, the Nation, and the world by accelerating
the development of new computing capabilities that can
transform science. When I last appeared before this
Subcommittee in May of 2013, we testified about the importance
of funding the development of exascale computing and the
dangers to U.S. leadership in computational science if we fail
to move expeditiously. Since that time, the urgency has
increased, as has our knowledge of how to proceed.
In February 2014, ASCAC reported to DOE on the top 10
exascale research challenges. This report reflected the
progress since our earlier 2010 exascale report. In addition to
identifying the 10 challenges, our expert panel emphasized both
that the United States has the technical foundation to address
and overcome them and that it is critical that we do so.
In August 2014 the Secretary of Energy Advisory Board Task
Force on Next-Generation Computing, of which I was a
participant, made public in its draft report, which included
the recommendation that DOE move forward with next-generation
computing at the exascale level. The report also endorsed
continued use of the co-design process and of government-
industry-academic partnering mechanisms. ASCR, in collaboration
with the National Nuclear Security Administration, has
developed the preliminary plan for such an exascale computing
initiative. This plan was provided to ASCAC for review last
November. This review is actively in process with the resulting
report due in September 2015 and an interim report at the end
of March.
I think it is more important than ever for the United
States to maintain and extend its leadership in scientific
computing. I hope that our presence here today will help to
that end. Thank you very much.
[The prepared statement of Dr. Giles follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman Weber. Thank you, Dr. Giles.
And, Mr. Turek, you are now recognized for five minutes.
TESTIMONY OF MR. DAVID TUREK, VICE PRESIDENT,
TECHNICAL COMPUTING, IBM
Mr. Turek. Good morning, Chairman Weber, Ranking Member
Johnson, and Members of the Subcommittee. Thank you for the
opportunity to speak with you about the Office of Science ASCR
program, supercomputing, and American technology leadership.
I have been involved in many of IBM's activities and
supercomputing over the last 25 years. During that time, I have
worked closely on supercomputing projects with both the Office
of Science and NNSA such as ASCI White, Blue, and Purple
systems at Lawrence Livermore; the Blue Gene systems, Mira, and
Sequoia at Argonne and Livermore respectively; the Roadrunner
system at Los Alamos; and as well as key software projects at
Pacific Northwest National Lab. I have witnessed firsthand the
magnitude of innovation possible courtesy of the collaboration
between private industry and the national labs.
I would like to pose today three questions with respect to
the linkage between supercomputing and technological
leadership. First, why be concerned about supercomputing
leadership? The Council on Competitiveness has stated that to
out-compete you must out-compute. I believe this to be true.
Supercomputers, as the other panelists have said, are tools for
inside strategic advantage with broad and diverse application
in areas such as oil discovery, fraud detection, efficient
automobile and aerospace design, and even many areas of basic
science. It is nearly axiomatic that better supercomputers give
one a chance for more insight and greater advantage than those
with lesser supercomputers. That is why you see the Europeans,
the Chinese, the Japanese, and others making a concerted push
through public funding of major supercomputer projects. They
want to out-compete us.
But there is a fundamental understanding we must also have.
Supercomputers are nothing without the software programs and
applications that run on them and software engineers only want
to produce software for the best machine, not the second,
third, or fourth best. Without the best supercomputers
available in the United States, software developers will
migrate to develop their innovations elsewhere. Once that trend
starts, it is very hard to stop or reverse. It is much more
costly to catch up than it is to stay ahead.
The second question is what technology problems are in the
way of maintaining leadership? The first problem is the need to
make supercomputers more energy efficient. The fastest Western
economy-based supercomputers in the world today consume about
10 megawatts of energy or $10 million a year. As supercomputers
get bigger and more powerful, without some real breakthroughs,
by the beginning of the next decade the energy bill could
easily be 100 megawatts or $100 million to run. This means the
cost of energy will begin to overtake the cost of the computer
itself, that becoming a limiting factor in supercomputer usage.
A slowdown in usage will ultimately correlate with a slowdown
in innovation and impact economic competitiveness.
The second problem is how to handle huge amounts of data.
It is clear that the explosive growth of data is challenging
some of the fundamental design principles of supercomputers.
For example, 500 e-books is about a billion bytes of data. With
today's technology, that amount of data can be moved through a
computer network in a matter of minutes or less. But suppose we
multiplied that amount of data by a million? That would
represent the amount of data many supercomputers are working on
today and in short order there will be problems a thousand
times beyond that.
Old design principles don't solve this problem. We cannot
simply do what we did in the past at greater scale to fix this.
The temptation, therefore, would be to ignore portions of data
to make the problem more tractable, but data left unanalyzed is
insight undiscovered, so we have to find ways to make future
supercomputers more accommodating to the vast amounts of data
they will be asked to explore. New innovations are requiring
networking, memory design, storage innovation, and data
management software to remedy this circumstance.
The third problem is application software. Most application
software running on supercomputers today are based on
mathematical approaches more than 40 years old, which is the
last time there was a major systematic government investment in
new algorithms. The software is now horribly mismatched to
modern supercomputers simply because 40 years ago no one could
have guessed what today's supercomputers would look like.
Access to modern software and new algorithms will have a
dramatic impact on the utility of modern supercomputers. There
must be a plan to modernize application software. There is no
silver bullet to solve these problems. Inventions required to
maximize impact, all the problems must be addressed in concert.
The third question is what needs to happen to maintain
leadership? From my experience, collaboration with the national
labs has been a proven means to stimulate innovation in
supercomputers. The labs work on problems of such complexity
they always stretch the limits of computing technology. In
fact, a crude rule of thumb is the computing requirements of
the national labs are about five to seven years advanced over
the rest of the market. Finding the ASCR program will present
the opportunity to address the problems I described and
contribute to maintain the pace of innovation competitiveness
demands. If this commitment is made, U.S. leadership in
supercomputing should be preserved for years to come.
Thank you very much and I would be happy to answer your
questions.
[The prepared statement of Mr. Turek follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman Weber. Thank you, Mr. Turek.
And now, Dr. Crowley, you are recognized for five minutes.
TESTIMONY OF DR. JAMES CROWLEY, EXECUTIVE DIRECTOR,
SOCIETY FOR INDUSTRIAL AND APPLIED MATHEMATICS
Dr. Crowley. Good morning, Chairman Weber, Ranking Member
Johnson, and Members of the Committee.
As noted in my introduction, I am Executive Director of the
Society for Industrial and Applied Mathematics, or SIAM. SIAM
comprises over 14,000 members who work in industry, government
and national labs, and in academia. They represent over 500
universities, corporations, and research organizations from
around the world. SIAM is dedicated to solving real-world
problems through applied mathematics and computational science.
Thank you very much for allowing me to testify and for
highlighting the critical work of the Department of Energy's
Office of Science and its Advanced Scientific Computing
Research program. SIAM greatly appreciates your Committee's
continued leadership on, and the recognition of, the critical
roles of the Office of Science and ASCR in enabling a strong
U.S. economy, workforce, and society through mathematical,
scientific, and engineering research relevant to the DOE
mission.
The Office of Science supports basic research to address
pressing challenges in energy, computing, physical sciences,
and biology and this support has been critical to the applied
mathematics and computational science community.
I wish to focus on three topics: ASCR support for
mathematical and computational science research, the potential
benefits of exascale and the technological challenges to reach
it, and finally workforce and training needs. First, the role
of ASCR in supporting key mathematical and computational
research.
ASCR supports the development of new modeling simulation
and data tools to help researchers solve scientific and energy
challenges. Modern life as we know it, from search engines like
Google to the design of modern aircraft, would not be possible
without the unique contributions of mathematicians and
computational scientists. Likewise, DOE depends on mathematical
and computational techniques to make predictions, model and
simulate systems that would be costly or impossible to
experiment on, and manage and make sense of ever-growing data
that is produced by scientific experiments such as DOE's
particle accelerators and light source facilities.
The Nation faces critical challenges in energy efficiency,
renewable energy, future energy sources, and environmental
impacts of energy production and use. These challenges all
involve complex systems such as the power grid or the U.S.
nuclear stockpile. Mathematical and computational tools help us
model and understand these systems, design new solutions to
problems, and predict the impact of new technologies. ASCR
programs not only support new mathematical tools but also
develop software so that DOE, industry, and the academic
community can use these tools. And I note that the PETSc team
at Argonne just was awarded the ACM SIAM prize in computational
science and engineering and that shows the power of the people
working at DOE.
Second, I would like to address the possibilities and
challenges of exascale. For all the advances that ASCR has
already enabled, today, there are still challenges that are too
complex for current computers to model. Exascale computing has
the potential to spur revolutionary advances in modeling and
simulation, expand our capacity to analyze complex systems in
great detail, and capture more complexity with better
predictive abilities than ever before.
I will note that the investments in modeling, algorithm
research, and software development are essential to realizing
the full benefits of exascale computers so that we can use
these machines to solve pressing scientific and energy
challenges. It is not just the hardware; the computer science
and the math are essential.
Finally, I would like to discuss an important workforce
development program within ASCR. Researchers trained to use
high-performance computers to solve key scientific challenges
are central to DOE's mission. The Computational Sciences
Graduate Fellowship program is a critical program that
maintains the pipeline of this workforce by supporting the
training of new scientists and engineers with strong
computational research experience and close ongoing ties to DOE
and the national labs. The CSGF has a long history of success
at DOE and SIAM strongly supports its continuation.
I thank you again for the opportunity to provide this
testimony today and I am happy to answer any questions. I have
provided additional details in my written testimony. Thank you.
[The prepared statement of Dr. Crowley follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman Weber. Thank you. I thank the witnesses for their
testimony. Members are reminded that the Committee rules limit
the questioning to five minutes and the Chair recognizes
himself for five minutes.
Good grief, where do we start? You all have just raised a
whole bunch of questions. Dr. Crowley, can you provide an
overview in plain English so our constituents can understand?
You kind of went through it there toward the end of what ASCR
program does but why is it important to the U.S. economy?
Dr. Crowley. The tools that are provided for modeling and
simulation are used across--I gave you the example of the
award--the prize that went to the PETSc team at Argonne
National Lab. PETSc is a team that has developed computational
tools for high-performance computers. These tools are used by
industry that do modeling and simulation for advancing
materials, some of the things that Roscoe Giles mentioned, and
without those tools, one can't use the computers efficiently to
do that. And so it is the research into not only developing the
tools that the people can use the computers but also the models
that run on them.
I mean to take an example that is not necessarily a DOE
model but just looking at one that came home to me recently
because of the weather prediction in Philadelphia that almost
kept me from coming here, this latest snowstorm missed
Philadelphia but it was predicted to dump more than a foot of
snow on us. Improved modeling, better tools, and higher
performance computing would have made that ability to make
those predictions much better. And that same thing applies for
any other kind of thing that is modeled across science and
engineering, that with better tools for modeling, better
computational tools, we can advance our ability to produce
better materials, to simulate anything that we need--and
understand better scientific things in fusion or in any other
area.
Chairman Weber. Okay. I think it was you that said that the
algorithms or the math that was used 40 years ago--it was Mr.
Turek so this question is probably for you--couldn't predict or
you couldn't see what computers look like today. What did you
say about that?
Mr. Turek. Yeah, what I meant by that was that at that time
frame, the nature of what was considered to be a supercomputer
bears no resemblance to the kinds of computers that exist
today, so people designed the algorithms and the corresponding
software to map to that kind of computer. Those approaches
don't translate well over four decades to the kinds of things
we are doing today.
Chairman Weber. So here is my question. Do we have the
capability today to look out 40 years in the future and predict
how effective those algorithms will be or will there be new
techniques? With advanced computing today are we able to look
out 40 years in the future?
Mr. Turek. Nobody can look out 40 years correctly. However,
what I would say is that we know that many of the algorithms
and the software implementations today are obsolete for what we
are trying to do. The way to characterize it would be the
following: Today's modern supercomputers typically use order of
millions of microprocessors. Many of the algorithms and the
software implemented only scale to maybe a handful of hundred
of microprocessors not because it can be done; it is because it
is a byproduct of the fact that that invention is 40 years old.
A reinvestment in algorithmic development, the fundamental
mathematics and the associated software, has been demonstrably
proven in places like Argonne, Lawrence Livermore, and Oak
Ridge that these approaches to common problems can be modified
to accommodate this nature of supercomputing we have today. You
would have a material effect on dramatically improving the
insight that people gain from the application of the
supercomputing tool.
Chairman Weber. Is part of the aim of ASCR, for example--
because we hear a lot in today's society about hacking and so
we invest the money and I am a great believer that we need to
be on the cutting edge because it helps national security, for
example, but are we at risk with supercomputing of investing
money, time, and resources, and then having that technology
stolen from us by other countries
Mr. Turek. So there is this notion of internationalism if
you will, but I would characterize it this way: The Chinese
program is very parochial to China. The European program is
very parochial to Europe and they are making investments that
are very much wedded to the parochial interests of companies
and institutions in those geographies. There is always the
chance that through regular commerce or more nefarious means
technology can escape geographic boundaries, but I think the
deployment of technology in the economy is what really makes a
difference, so the more supercomputing that can be made
available, the more and diverse kinds of people who can get
access to it and use it is what really spurs the economic kind
of innovation we have all alluded to here today.
Chairman Weber. Yeah. Well, I appreciate that. And I am out
of time so the Chair will now recognize Ranking Member Johnson.
Ms. Johnson. Thank you very much, Mr. Chairman.
I am so delighted we have such able witnesses today and I
know that this hearing is focused on our investments in
supercomputing research in particular, but I would like to take
advantage of your presence, Mr. Augustine, to ask a few broad
questions to help us guide the future in how we are able to
continue research, whether or not we are producing the
researchers. In 2005 the National Academies' Gathering Storm
Panel, which you chaired, recommended increasing science agency
budgets by ten percent annually.
The 2007 COMPETES bill, which was very graciously accepted
and supported by President Bush, had bipartisan support for a
positive growth trajectory of R&D, and unfortunately,
appropriations for the last eight years have not come close to
keeping up with what was projected. It was changed to a more
conservative recommendation to at least four percent annually
in 2014.
In the current budgetary and political environment, how
would you continue to make the case for increased funding for
R&D to politicians across the political spectrum? And what do
you believe are the consequences if we do not even achieve this
modest four percent annual growth target for federal investment
in basic research and development? And, finally, do you believe
that a robust reauthorization of America COMPETES should be a
top priority for this Committee this year?
Mr. Augustine. Well, thank you for that question.
Chairman Weber. Mr. Augustine, turn your mike on, please.
Mr. Augustine. I thought it was on.
Chairman Weber. Oh, there you go.
Mr. Augustine. Sorry. To deal with the last part of your
question first, I think America COMPETES is perhaps the most
important thing that this Committee could take on. It drew more
attention to the problems we face in this area and took further
steps to improving the situation than anything else I am aware
of that we have done. So I would strongly urge that.
With regard to the status of the research and where we have
come since the various reports that you allude to, the bad news
is that we are declining in our investment in research as a
percentage of GDP. Other countries are growing. Even at NIH,
which is--research there is strongly favored by the American
public--we have seen a 22 percent cut in the last decade in
real dollars and it is continuing to decline. This of course
discourages young people from going into research and basically
it means that we are going to have a lower quality of life,
impact on our health will be very real, and the economy today
is so heavily dependent on technology that without doubt we
will be hurt economically seriously.
I would cite an example from my own field of the impact of
research and particularly high-performance computing. I am an
aerodynamicist, design airplanes, among other things. The way
we used to design airplanes when I was early in my career was
built giant wind tunnels. We built them when they were plugged
into the Tennessee Valley Authority by and large because that
was the only place we could get enough power. We ran them at
night we didn't shut down the lights in the southern part of
the country.
Today, we don't use wind tunnels. We put the airplane and a
high-performance computer if you will, use a mathematical model
and within a nanosecond have the answers that we are
researching, just one example of the enormous impact that
investment in technology can have and also the negative impact
of not investing in science, research, and technology.
Ms. Johnson. Well, thank you very much.
The National Research Council report entitled ``Rising to
the Challenge: U.S. Innovation Policy for the Global Economy,''
states the assumption that the output of the U.S. innovation
process will be captured by U.S.-based industry has been
rendered obsolete by globalization, and that knowledge created
through federally funded research at universities and national
laboratories can be commercialized and industrialized virtually
anywhere. The report goes on to say that a more comprehensive
innovation policy is needed to anchor new and existing
companies here in the United States.
The American Academy of Arts and Sciences panel that you
recently chaired addressed some of this issue in a report
released this fall. What recommendations do you have for what
federal policies are necessary to ensure that U.S. companies
benefit from U.S. innovation?
Mr. Augustine. Well, thank you for that question. And as
you point out, research is a global commodity or global asset,
and it raises a question why not just let others do the
research and then apply their research? The answer, I would
cite Craig Barrett, who ran Intel some years ago. Craig says
that on the last day of any calendar year 90 percent of the
revenues that Intel receives are for products that didn't exist
on the first day of the calendar year, and so the only answer
to your question that I can see is that we just have to be
faster than others in applying the results of research. We have
got to be fast.
And your question what do we do about it and the answer is
remove every bureaucratic obstacle, every obstacle we can think
of, particularly in technology transfer from the labs, that
causes time delay because time is everything.
Ms. Johnson. Thank you very much. My time has expired.
Thank you.
Chairman Weber. Thank you.
And the Chair now recognizes the Vice Chairman of this
Committee, Congressman Newhouse.
Mr. Newhouse. Thank you very much, Mr. Chairman. I
appreciate that and appreciate you gentlemen being here this
morning and talking about this very important subject. It is
certainly enlightening me as to the nature of our
responsibility here.
Not to let you dominate the program this morning, Mr.
Augustine, but a question that arose in my mind after reading
through your testimony that a lot of the body of research at
our national laboratories is maybe not being utilized as much
as it could be so to speak, not to put words in your mouth, but
there are certain obstacles that stand in the way of getting
that research to industries. So could you talk a little bit
about maybe what you see as solutions to that issue that we
have? Is it communication, some of the conflict-of-interest
issues that you mentioned, and those kinds of things?
Mr. Augustine. Well, thank you, Congressman Newhouse, for
that question. And I do believe that the Nation doesn't begin
to benefit from the asset that our national labs represent. It
certainly benefits importantly but it could be so much more,
and the reason for that is that we need to do a better job of
getting knowledge out of the laboratories and into industry so
that we can commercialize and distribute the results.
And as to impediments, there are many. One that certainly
stands in my mind is that firms simply don't know what is going
on in the national laboratories. They tend to be rather
isolated. And we could do a much better job of letting people,
industry, know what is happening at the laboratories.
Secondly, the best way to transfer technology that I have
ever been able to find is by transferring people. You move the
knowledge that is in their minds. And today, well-meaning
conflict-of-interest laws make it very difficult to transfer
people among industry, government, and academia. In my career I
had the opportunity to put in two tours in government and today
I doubt that I could do that under the conflict-of-interest
laws that exist.
A third one that I would cite is that we are very
concerned, properly so, about favoring one firm over another.
What do we do about it? Without taking a great deal of time,
one is for the labs to do a better job of letting the world
know what they are working on, the industrial world if you
will. Other things that are cited in H.R. 5120, for example,
giving the labs more latitude to create industry partnerships,
give the labs more latitude to negotiate technology transfer
agreements. These are a few of the things that could be done
but I don't have answers to the conflict-of-interest one
because obviously we don't want conflicts of interest. On the
other hand, the inability to move people and to move ideas in
and out of the labs is a huge burden on our country.
Mr. Newhouse. Thank you. I appreciate that.
Mr. Augustine. Thank you.
Mr. Newhouse. Quickly, a question then perhaps for Mr.
Turek and perhaps Dr. Giles as well. It is--my limited
understanding is that the largest supercomputers are rarely
able to operate at full capacity due to their complexity, some
components almost always in need of attention or repair. If
that is a true statement, could you tell me what is being done
to improve the reliability of these systems and are we devoting
enough resources to this aspect of advancement?
Mr. Turek. I will take the first shot at it. We are doing a
lot for that. A lot of that is actually handled by software so
soft recoveries of problems. What you see with supercomputing
are problems of scale. If you have a million parts of anything,
the likelihood is you are going to see something failing pretty
regularly, even if it is integrated circuits. It is a problem
that has been understood for quite some time and principally is
handled by software techniques to overcome it. So in the vast
majority of cases you actually can get to full capacity if you
have the software capability on the application level to
utilize it. That is the bigger impediment right now. Again,
most people who gain access to commercial software are gaining
access to software that is archaically designed relative to the
scale of the kinds of computers being built today and that is
the limiting factor.
Dr. Giles. Can I add something?
Mr. Newhouse. Absolutely, Dr. Giles.
Dr. Giles. I think that--yes, I think that also our sense
of what the capacity of a system is reflects some of the
archaic history in the sense that we often measure or think of
a capacity is how much data can you sort of crunch, transform
from one form to another, which is an artifact of the time when
the critical component of a computer was the processor that
made that transformation. Now, people are looking at systems
with millions of processors and redundancy in processors is not
a negative to have multiple processors comparing results one to
another. So, as Mr. Turek said, there are lots of opportunities
for new ways of ensuring the reliability of the final answers
we get.
And if we get discouraged about thinking about that
problem, I would remind us all that our brains, with millions
and millions of--and billions of neurons and interconnections
have faults on the neuron level all the time and they don't
materially affect the ultimate outcome, and I think we are in
the process of building computers that can function more like
that.
Mr. Newhouse. Thank you very much.
Thank you, Mr. Chairman.
Chairman Weber. Thank you.
And the Chair now recognizes Congressman Hultgren from
Illinois for five minutes.
Mr. Hultgren. Thank you all so much for being here. Thank
you, Chairman. I especially want to thank the Chairman for
working out a way for Dr. Giles to be with us remotely.
I am very fortunate to represent Fermilab and I have
Argonne right down the road from me. Because of this, I have
been able to see the fruits that grow out of our Nation's
commitment to basic curiosity-driven scientific research. The
impacts of this research I believe are limitless. Just as we
didn't go to the moon to invent Velcro, we didn't build
particle colliders so that we could invent the magnet for our
MRI machines.
This topic, supercomputing, is close to home for me because
physics is where big data began. Besides the maintenance of our
nuclear stockpile, it is either astrophysics or high-energy
physics that is driving the research necessary to build the
most sophisticated computer networks we have today. Because of
this, it was largely DOE that began the genome project before
NIH realized it was a feasible endeavor. As interested as I am
in technology transfer and local economic development, if our
research enterprise is focused on the short-term photo op and
press release-style research, which it appears the
Administration is more prone to advance, we will lose out on
the long-term benefits we all say we should be focused on. If
we are going to stay at the forefront of technology or
technological development, we must reaffirm our commitment to
basic scientific research.
Dr. Giles, in our previous hearing, you had a chance to
review a draft copy of my legislation, which in the 113th House
eventually passed, H.R. 2495, the American Supercomputing
Leadership Act. My bill called for a lab-industry-university
partnership to develop two different exascale machines. I
wondered if you would be willing to describe what industry's
role should be in such a partnership and then describe the
benefits of having a university as part of this partnership?
Dr. Giles. Yes, I would be happy to address that and some
of my written testimony does get to that point. I think that
ASCR's work has helped to start a virtuous cycle with industry,
academia, and the labs in developing and looking forward to the
path for exascale so that in collaboration with industry we are
able to have government funds help to stimulate research and
investigation in areas that are important for building the next
generation scale of computers before that is actually
competitive or something that is in the competitive spirit of
the industry, but then industries impact is to help define what
is sufficiently along the lines of work that they can build and
build on into something that they would be interested in from
their perspective, that we find an accommodation.
In the co-design methodology that I mentioned represents
the pattern of developing new software and algorithms as--in
the context of hardware that is evolving and to help use those
needs from the scientific community, from the universities and
the labs to help define what kind of hardware makes sense so
that the--this goes back to the idea of building an ecosystem
that supports rapid advances in scientific computing that links
together all those elements.
I do want to thank you so much for the legislation you
propose that we discussed last time and which made it out of
the House, as I understand it, but not all the way through the
end of the process. You know, I think it is a really important
step that we explicitly fund the development of that next
generation better systems.
Mr. Hultgren. Thanks, Dr. Giles.
Quickly, Mr. Augustine, I would first like to thank you for
all of your work. You have been a leader in this and in so many
other spaces, it is amazing. Thank you.
I had the pleasure of sitting down with your colleague Dr.
Neal Lane to discuss local economic development potential for
the national labs in reference to the Restoring the Foundation
report. Many of the recommendations from this discussion echoed
my previous passed legislation, the DOE Labs Modernization and
Technology Transfer Act, which the Bipartisan Policy Center
listed in their doable items, which there aren't too many of,
for the 114th Congress. I wonder if you could make a comment
more generally on this bill and the needs and benefits for
making the labs more nimble and open to the public?
Mr. Augustine. Well, yes. One of the things that certainly
relates to what you raise is that the labs are able to build
major facilities that individual firms can't afford to build.
Fermilab is a classic example. And if they are not available to
the public or industry by and large, then we don't begin to get
the value from them that we could get. Some of the legislation
that you describe takes important steps in this regard.
I guess I would say in terms of a broad answer--and I
realize that we are running out of your time--that the bad news
is that we spend, as I said, a 10th of a percent of the GDP on
research. The good news is you could double that and only have
to allocate a 10th of a percent of the GDP. And so the
opportunity is probably there to make major changes.
I go back to one of the studies that you refer to. We
discovered that we spend more on potato chips in this country
than we spend on research on clean energy. That just doesn't
make sense.
Mr. Hultgren. Well, again, I want to thank you all for
being here.
Thank you, Chairman.
And real quick, just thank you, Dr. Crowley, too, for the
shout-out to Argonne and the recent recognition there. That is
fantastic. So thank you so much.
Chairman, I yield back.
Chairman Weber. Thank you, sir.
The Chair now recognizes Mr. Massie from Kentucky.
Mr. Massie. Thank you, Mr. Chairman.
My question is really for anybody up there that cares to
comment, but it seems like 20 years ago there was the
apocryphal prediction that we would run out of available
computing power with silicon, yet here we are still on silicon.
What is the next step after silicon? And since we didn't run
out of power with silicon how much further can we go on
silicon?
Mr. Turek, it looks like you are interested in answering
that.
Mr. Turek. I will take the first shot at least.
We are at an apocryphal time and to a certain extent you
could characterize the industry as putting a Band-Aid over this
problem. So the limitations of silicon are embedded in physics.
We are at those limits today. I think the last time I saw an
advertisement on TV about buy a computer because the processor
is faster was January 2001. You don't have a 10 gigahertz
processor. You are never going to see one either because the
physics are limiting.
So instead what the industry has done is it has spewed out
massive amounts of cores, lower-power compute elements that are
ganged together to work in concert on the problems at hand. The
problem is you don't get a linear scalability of the compute
effect. So in other words, if I have four cores, I don't get
four times the compute capability of one core. Maybe I get 2.5.
And as I scale up to a million, I am not getting a million
times; I am getting something far less than that.
So we are Band-Aiding our way through this limitation at
the physics level. There are more materials and so on that are
coming forth and whether it is carbon nano tubes or something
else, but physics is a limiting factor here.
The way you deal with this ultimately is you look at the
architecture of how these systems are put together and the
composite set of technologies that let you deal with the
problem. Advances in networking technology, memory systems, all
these things need to be looked at in total to begin to push the
ball forward but it is the real slog now. Believe me, in 1996 I
knew how to build a Roadrunner system, not a problem; it was
just a matter of hard work. That was the first petascale system
on the planet. In 2005 I didn't know how to get to exascale and
still struggle today. We are up against real limits.
Mr. Massie. So does anybody else care to talk about that?
Dr. Giles. Yes, just to add one quick observation. The
Secretary of Energy Advisory Board Task Force considered very
seriously this question about the relationship of what we are
doing now to--for the future, and one of the things that became
very clear is that because the limitations and the
possibilities and opportunities are physics-based and the DOE
labs are the premier research set of facilities for the
physical sciences, that in some ways DOE with its computing
interest and capability and the labs is in an excellent
position to do the research needed to move beyond silicon and
CMOS and what we are doing now to the next generation, whether
that involves, as David said, superconducting technology or
quantum technology, the labs are in a really good position to
investigate.
Mr. Massie. That was going to be my next question. So
obviously we have already hit the physical limits of silicon
and the speed of light and energy density and all that stuff,
and we have Band-Aided that with architecture or maybe that is
the way around it, but we have diminishing returns to putting
more cores in there. What are the next promising platforms and
what role will our research that we are paying for here in
Congress play? What is the next transistor? What is going to be
the next paradigm shift and what role does our research play in
that?
Mr. Turek. Well, I will make a brief comment. There is no
silver bullet. There is nothing I can point to that says the
problems of the future are done; we can simply move along as
systematically as we have over the last 50 years or so. When I
talk about architecture I mean different approaches to solve
the problem.
Today, one of the techniques that is being explored and
reflected in the CORAL program at the DOE is the employment of
accelerators, specialized processors attached to conventional
processors to give an overall speed-up in compute capability.
We pioneered this, by the way, with a cell processor at Los
Alamos ten years ago, which was an accelerator-based kind of
technology. That is a new idea. Accelerators have been thought
of over many years but never gained acceptance because we could
leverage the evolution of silicon to overcome the limits. No
longer possible, now there is an embrace of accelerators. So
you see a lot of different kinds of accelerators come into play
and applied in very unique and interesting kinds of ways.
Mr. Massie. Thank you very much. I am excited to see what
the next breakthrough is. I realize there is no silver bullet
and we have got to use a shotgun, but I trust that we will come
up with something. Thank you.
Chairman Weber. Thank you. And I thank the witnesses for
their valuable testimony and the Members for their questions.
The record will remain open for two weeks for additional
comments and written questions from the Members.
So thank you, gentlemen. Thank you, Dr. Giles. The
witnesses are excused and the hearing is adjourned.
[Whereupon, at 10:07 a.m., the Subcommittee was adjourned.]
Appendix I
----------
Answers to Post-Hearing Questions
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
[all]