[Senate Hearing 109-1150]
[From the U.S. Government Publishing Office]



                                                       S. Hrg. 109-1150
 
                       HIGH-PERFORMANCE COMPUTING

=======================================================================

                                HEARING

                               before the

      SUBCOMMITTEE ON TECHNOLOGY, INNOVATION, AND COMPETITIVENESS

                                 OF THE

                         COMMITTEE ON COMMERCE,
                      SCIENCE, AND TRANSPORTATION
                          UNITED STATES SENATE

                       ONE HUNDRED NINTH CONGRESS

                             SECOND SESSION

                               __________

                             JULY 19, 2006

                               __________

    Printed for the use of the Committee on Commerce, Science, and 
                             Transportation



                  U.S. GOVERNMENT PRINTING OFFICE
71-158                    WASHINGTON : 2011
-----------------------------------------------------------------------
For sale by the Superintendent of Documents, U.S. Government Printing Office, 
http://bookstore.gpo.gov. For more information, contact the GPO Customer Contact Center, U.S. Government Printing Office. Phone 202ï¿½09512ï¿½091800, or 866ï¿½09512ï¿½091800 (toll-free). E-mail, [email protected].  


       SENATE COMMITTEE ON COMMERCE, SCIENCE, AND TRANSPORTATION

                       ONE HUNDRED NINTH CONGRESS

                             SECOND SESSION

                     TED STEVENS, Alaska, Chairman
JOHN McCAIN, Arizona                 DANIEL K. INOUYE, Hawaii, Co-
CONRAD BURNS, Montana                    Chairman
TRENT LOTT, Mississippi              JOHN D. ROCKEFELLER IV, West 
KAY BAILEY HUTCHISON, Texas              Virginia
OLYMPIA J. SNOWE, Maine              JOHN F. KERRY, Massachusetts
GORDON H. SMITH, Oregon              BYRON L. DORGAN, North Dakota
JOHN ENSIGN, Nevada                  BARBARA BOXER, California
GEORGE ALLEN, Virginia               BILL NELSON, Florida
JOHN E. SUNUNU, New Hampshire        MARIA CANTWELL, Washington
JIM DeMINT, South Carolina           FRANK R. LAUTENBERG, New Jersey
DAVID VITTER, Louisiana              E. BENJAMIN NELSON, Nebraska
                                     MARK PRYOR, Arkansas
             Lisa J. Sutherland, Republican Staff Director
        Christine Drager Kurth, Republican Deputy Staff Director
             Kenneth R. Nahigian, Republican Chief Counsel
   Margaret L. Cummisky, Democratic Staff Director and Chief Counsel
   Samuel E. Whitehorn, Democratic Deputy Staff Director and General 
                                Counsel
             Lila Harper Helms, Democratic Policy Director
                                 ------                                

      SUBCOMMITTEE ON TECHNOLOGY, INNOVATION, AND COMPETITIVENESS

                     JOHN ENSIGN, Nevada, Chairman
TED STEVENS, Alaska                  JOHN F. KERRY, Massachusetts, 
CONRAD BURNS, Montana                    Ranking
TRENT LOTT, Mississippi              DANIEL K. INOUYE, Hawaii
KAY BAILEY HUTCHISON, Texas          JOHN D. ROCKEFELLER IV, West 
GEORGE ALLEN, Virginia                   Virginia
JOHN E. SUNUNU, New Hampshire        BYRON L. DORGAN, North Dakota
JIM DeMINT, South Carolina           E. BENJAMIN NELSON, Nebraska
                                     MARK PRYOR, Arkansas


                            C O N T E N T S

                              ----------                              
                                                                   Page
Hearing held on July 19, 2006....................................     1
Statement of Senator Cantwell....................................     2
Statement of Senator Ensign......................................     1

                               Witnesses

Burt, Stanley K., Ph.D., Director, Advanced Biomedical Computing 
  Center.........................................................    46
    Prepared statement...........................................    49
Garrett, Michael, Director, Airplane Performance, Boeing 
  Commercial Airplanes...........................................    38
    Prepared statement...........................................    40
Jehn, Christopher, Vice President, Government Programs, Cray Inc.    21
    Prepared statement...........................................    23
Lombardo, Joseph M., Director, National Supercomputing Center for 
  Energy and the Environment, University of Nevada, Las Vegas....    35
    Prepared statement...........................................    37
Szykman, Dr. Simon, Director, National Coordination Office for 
  Networking and Information Technology Research and Development.     4
    Prepared statement...........................................     6
Waters, Jack, Executive Vice President/Chief Technology Officer, 
  Level (3) Communications.......................................    25
    Prepared statement...........................................    32
Wladawsky-Berger, Dr. Irving, Vice President, Technical Strategy 
  and Innovation, International Business Machines Corporation 
  (IBM)..........................................................    14
    Prepared statement...........................................    16

                                Appendix

West, Tom, CEO, National LambdaRail, prepared statement..........    61
Response to written questions submitted by Hon. John Ensign to:
    Stanley K. Burt, Ph.D........................................    70
    Michael Garrett..............................................    69
    Christopher Jehn.............................................    65
    Joseph Lombardo..............................................    68
    Dr. Simon Szykman............................................    63
    Jack Waters..................................................    67
    Dr. Irving Wladawsky-Berger..................................    64


                       HIGH-PERFORMANCE COMPUTING

                              ----------                              


                        WEDNESDAY, JULY 19, 2006

                               U.S. Senate,
       Subcommittee on Technology, Innovation, and 
                                   Competitiveness,
        Committee on Commerce, Science, and Transportation,
                                                    Washington, DC.
    The Subcommittee met, pursuant to notice, at 11:05 a.m. in 
room SR-253, Russell Senate Office Building, Hon. John Ensign, 
Chairman of the Subcommittee, presiding.

            OPENING STATEMENT OF HON. JOHN ENSIGN, 
                    U.S. SENATOR FROM NEVADA

    Senator Ensign. I call the Subcommittee to order. I want to 
welcome everybody to today's hearing on high-performance 
computing, and I want to thank my colleague, Senator Cantwell 
from Washington State, who was the inspiration for this 
hearing. I am very excited to listen and learn today.
    In an increasingly interconnected global economy, high-
performance computing plays an important role in maintaining 
the United States' economic and scientific competitiveness and 
national security. From aircraft and automotive design to 
weather prediction to advanced medical research to financial 
modeling on Wall Street, high-performance computing accelerates 
the innovation process by shrinking time to insight and time to 
solution for both discovery and invention.
    In the 21st century, together with theory and 
experimentation, computational science now constitutes the 
third pillar of scientific inquiry. High-performance computers 
enable researchers to build and test models of complex 
phenomena such as multi-century climate shifts and multi-
dimensional flight stresses on aircraft. Without high-
performance computers, these phenomena cannot be replicated 
effectively in the laboratory.
    High-performance computers also enable organizations to 
manage huge volumes of data rapidly and economically. As the 
Council on Competitiveness recognized at its High-Performance 
Computing Users Conference last year, companies that leverage 
high-performance computing tools realize a range of competitive 
benefits from shortened product development cycles and faster 
time to market to reduced research, development, and production 
costs, all of which improve a company's bottom line and the 
country's competitiveness.
    Moving forward, high-performance computing will continue to 
facilitate innovation in our Nation's industries, improve our 
research capabilities, and enhance our national security. For 
example, high-performance computing holds the promise of 
recovering 75 percent or better of an oil reservoir's capacity, 
up from 50 percent today, through more accurate seismic 
modeling.
    In addition, high-performance computing can help the United 
States to explore and maximize usage of alternative energy 
technologies, such as hydrogen fuel cells. High-performance 
computing can help optimize the design of a wide range of 
consumer products that we use every day, from cars to Pringle's 
potato chips, to make sure they are safe and reliable.
    High-performance computing can also support breakthroughs 
in medical research and treatments for disease. For example, it 
has been used to help figure out how Alzheimer's affects the 
brain. It has also been used to improve treatment for various 
forms of cancer.
    Finally, high-performance computing can assist in research 
undertaken across all scientific disciplines at our Nation's 
universities.
    I am eager to hear about the progress that is being made in 
high-performance computing by both the public- and private-
sectors. In addition, I look forward to discussing future 
challenges and opportunities that may impact the development 
and application of high-performance computing. I look forward 
to the expert testimony of our distinguished witnesses and I 
want to thank everyone for attending and participating in 
today's hearing.
    Before we hear from our witnesses, I'd like to welcome any 
opening statement by Senator Cantwell.

               STATEMENT OF HON. MARIA CANTWELL, 
                  U.S. SENATOR FROM WASHINGTON

    Senator Cantwell. Thank you, Mr. Chairman, and thank you 
for conducting this important hearing on high-performance 
computing, and thank you for your enthusiasm and interest in 
this particular area. I also want to thank the witnesses today 
and particularly the two with ties to Washington State. Michael 
Garrett, Director of Airplane Performance at the Boeing 
Commercial Aircraft Division, made a long trip from Everett, 
Washington, to join us today and he will describe how Boeing 
uses high-performance computers to design revolutionary 
aircraft such as the Boeing 787.
    I also want to welcome Mr. Christopher Jehn, Vice President 
for Government Programs from Cray Computers, based in Seattle. 
In my mind Cray is synonymous with high-capability, high-
performance computing, and Mr. Jehn is standing in for the CEO, 
who could not be with us today.
    This is an exciting time for high-performance computing in 
the State of Washington and across the country. Companies, 
universities, and research institutions are taking full 
advantage of our expertise in both high-performance computing 
and in life sciences. Nobel Prize-winning scientist Dr. Leroy 
Hood, Founder of the Institute of Systems Biology in Seattle, 
is leveraging the power of grid computing, for example, to 
accelerate the development of predictive, preventive, and 
personalized medicine.
    Meanwhile, the University of Washington, with one of the 
top computing science departments in the Nation, continues to 
apply cutting-edge research to advanced networking and 
distributed computer systems. And through an innovative 
collaboration with the Fred Hutchison Cancer Research Center, 
we are training the next generation of scientists in the 
nascent field of computational molecular biology.
    Other research institutions in the state are making use of 
high-performance computing assets through the Pacific Northwest 
Gigapop, and Microsoft is working to make high-performance 
computing technology more mainstream by developing software to 
network desktops and clusters of computer servers.
    Over the past decade, the high-performance computing 
industry has grown beyond the national security interests that 
first created it and drove it, and today researchers at 
universities, national laboratories and corporations with 
access to supercomputers are using their tremendous 
computational powers to model complex natural phenomena, to 
test expensive systems through simulation, and to replace 
experiments that are hazardous, illegal, or forbidden by 
official policies and treaties.
    So America has always been at a technologically-advanced 
stage when it comes to this kind of innovation. But now is not 
the time for us to fall behind, and to stay competitive as a 
Nation we must maintain computer leadership in high-performance 
computing and computational sciences.
    The U.S. Government still remains a primary user of high-
performance computing and we use it to maintain our military 
superiority, to achieve goals and to defend in other areas of 
national security. I also want to make sure that we continue to 
look at this important role as it relates to the Department of 
Energy and the various missions that the Department of Energy 
carries out.
    I look forward to hearing from many of the individuals here 
today, and particularly about some of those relevant 
applications as it relates to the Pacific Northwest National 
Laboratories in Richland, Washington.
    In 1991 when Congress passed the High-Performance Computing 
Act, the Act that was first established under the first 
President Bush, we had some goals and standards for Federal 
high-performance research. I think it is important now that we 
look at what kinds of changes and upgrades need to be made to 
that policy. That is why in 2004 I co-sponsored the High-End 
Computer Revitalization Act of 2004, which became law, and 
focused really only on the Department of Energy.
    So today I hope that we can discuss how we need to broaden 
that to focus on other areas. So I look forward to working with 
the Chairman as we move forward on this important issue of 
high-performance computing and the needs for our Nation and 
what other additional language, whether that is H.R. 28, the 
High-Performance Computing Revitalization Act of 2005, or other 
language that helps us maintain our effectiveness and our 
advantage in high-performance computing as a Nation.
    So, I thank the Chair.
    Senator Ensign. Thank you, Senator Cantwell.
    Our first witness is Dr. Simon Szykman. Dr. Szykman is the 
Director of National Coordination Office for Networking and 
Information Technology Research and Development, and we welcome 
your testimony and welcome you to the Subcommittee.

           STATEMENT OF DR. SIMON SZYKMAN, DIRECTOR,

        NATIONAL COORDINATION OFFICE FOR NETWORKING AND

        INFORMATION TECHNOLOGY RESEARCH AND DEVELOPMENT

    Dr. Szykman. Thank you very much, Mr. Chairman, Senator 
Cantwell. I am pleased to have been invited here today to 
discuss the role of the government in funding high-performance 
computing, or HPC, research and development.
    The Networking and Information Technology Research and 
Development Program, which I will refer to as the NITRD 
program, represents the coordinated efforts of many Federal 
agencies that support R&D in the areas of networking and 
information technology. I am the Director of the National 
Coordination Office for the NITRD program, the office which is 
responsible for supporting interagency technical and budget 
planning and assessment for the NITRD program.
    Today I would like to discuss three different aspects of 
high-performance computing: its place as a priority in the 
Federal Government R&D portfolio, the impact of successful 
interagency coordination in this area, as well as U.S. 
leadership in HPC technologies.
    As the NITRD program has evolved over the years, HPC has 
not only remained the dominant element of the program, but has 
been cited on a recurring basis as a priority within the 
Federal R&D portfolio. This has led to significant investments 
in HPC. In Fiscal Year 2002, funding for HPC in the NITRD 
program was less than $800 million. In 2007, next year's budget 
request, the budget has grown over 65 percent since then to a 
budget request of over $1.3 billion for HPC.
    Development of high-end computing capability and capacity 
is also a priority research area for the American 
Competitiveness Initiative which was announced by the President 
earlier this year. The Fiscal Year 2007 HPC budget requests for 
the ACI agencies--NSF, DOE's Office of Science, and NIST--are 
collectively over $135 million above 2006 levels. DARPA, 
although not part of the ACI, is also a strong supporter of HPC 
R&D and is expected to have a budget increase of $23 million 
next year.
    The release of the ``Federal Plan for High-End Computing'' 
in 2004 represented the start of a renewed emphasis on HPC R&D 
within the Federal NITRD program. Interagency coordination with 
strong leadership from the Office of Science and Technology 
Policy and OMB in the White House have resulted in 
unprecedented cooperation on HPC issues across the Federal 
Government.
    Programs such as DARPA's HPCS, High Productivity Computing 
Systems program, and the NSF-led High-End Computing University 
Research Activity have garnered support from all of the Federal 
agencies involved in HPC R&D within the NITRD program. More 
importantly, recognizing the importance of these efforts for 
next-generation technologies, several agencies are providing 
their own funding to support these programs in addition to the 
funding provided from DARPA and NSF.
    In other noteworthy policy developments, addressing the 
issue of accessibility of HPC resources DOE and NASA have 
opened up their resources to communities beyond their 
traditional research communities. This has enabled millions of 
hours of supercomputing time to be made available to industry 
projects as well as other government agencies that in the past 
would not have had access to these HPC resources. HPC system 
procurement practices are also being influenced through the 
sharing of best practices, performance metrics, and benchmarks 
developed through interagency collaboration.
    In 2002, HPC gained high visibility when Japan announced 
the bringing online of a new supercomputer called the Earth 
Simulator. Although this garnered some attention within certain 
policy circles, the government research community had been 
aware of this system being developed. three weeks ago, a new 
version of the Top500 Supercomputer Sites list was released. 
The list, which surveys the world's 500 fastest supercomputers, 
clearly confirms that the U.S. continues to hold a leadership 
position in HPC technologies.
    Some interesting statistics drawn from the latest version 
of the list: The Earth Simulator which I just mentioned now 
sits at the number ten position, not the number one position. 
Of the nine machines in front of it, six of them are inside the 
United States, including the top four machines. The U.S. 
dominates the list as a whole, with over 60 percent of the 
world's 500 fastest machines being in the United States. U.S. 
vendors are dominant suppliers of HPC technologies. The top 
three vendors account for approximately 75 percent of the 
world's 500 fastest systems. Even those outside of the United 
States rely strongly on U.S.-developed and U.S.-sold 
technologies.
    Looking back, we can confirm that the launch of the Earth 
Simulator did not represent a crisis for U.S. competitiveness 
in the context of HPC technologies. This is very important to 
note in the context of recent announcements from Japan 
indicating that they are undertaking the development of a new 
next-generation supercomputing system over the next few years 
as a successor to the Earth Simulator.
    The fact that the U.S. currently holds the title of world's 
fastest supercomputer does not herald a new era in HPC 
leadership any more than the loss of that number one position 
represented a loss of leadership several years ago. HPC has 
been and will continue to be a priority within the Federal R&D 
portfolio. The clearest demonstration of progress over the past 
4 years should not be viewed in terms of the raw speed of the 
world's fastest machine, but rather in the context of a growing 
focus on HPC technology policy in the government, unprecedented 
interagency coordination and collaboration on planning and 
implementation of plans within the government, and the 
increasingly cooperative ties between the government and the 
private sector.
    The progress that has taken place has been the result of 
concerted efforts aimed at fostering a vibrant government 
research community, as well as the work of many dedicated 
individuals working collaboratively toward shared objectives 
and goals.
    Once again, I would like to thank you for the opportunity 
to be here today and I am happy to answer any questions.
    [The prepared statement of Dr. Szykman follows:]

      Prepared Statement of Dr. Simon Szykman, Director, National 
Coordination Office for Networking and Information Technology Research 
                                  and 
                              Development

    Mr. Chairman and members of the Subcommittee, I am pleased to have 
been invited here today to discuss with you the role of the Federal 
Government in funding high-performance computing research and 
development (R&D), and to place these investments in the broader 
context of global competitiveness.
    The Federal Networking and Information Technology Research and 
Development (NITRD) Program, was established by the High-Performance 
Computing Act of 1991 (Pub. L. 102-194) and further elaborated upon by 
the Next Generation Internet Research Act of 1998 (Pub. L. 105-305). 
Federal networking and information technology research and development, 
which launched and fueled the digital revolution, continues to drive 
innovation in scientific research, national security, communication, 
and commerce to sustain U.S. technological leadership. The NITRD 
Program, now in its 15th year, represents the coordinated efforts of 
many Federal agencies that support R&D in networking and information 
technology.
    I am the Director of the National Coordination Office (NCO) for 
Networking and Information Technology Research and Development. The 
NITRD National Coordination Office is responsible for supporting 
technical and budget planning and assessment activities for the NITRD 
Program. The interagency coordination of NITRD activities takes place 
under the auspices of the National Science and Technology Council 
(NSTC), and more specifically through the NSTC's Networking and 
Information Technology Research and Development Subcommittee, and 
several interagency working groups and coordination groups that operate 
under this Subcommittee. The collaborative efforts of the interagency 
NITRD community increase the overall effectiveness and productivity of 
Federal networking and information technology R&D investments.
    Today I would like to discuss three different aspects of high-
performance computing: (1) high-performance computing as a priority in 
the overall Federal R&D portfolio, (2) the impact and success of 
interagency coordination in the area of high-performance computing, and 
(3) U.S. leadership in high-performance computing in the context of 
global competitiveness in information technology and its applications.
High-Performance Computing as a Priority in the Federal R&D Portfolio
    Fifteen years ago, what is now the NITRD Program was established in 
legislation as the National High-Performance Computing and 
Communications Program, having at that time a narrower focus on R&D in 
high-performance computing technologies and high-speed networks. Today, 
investments in high-performance computing support a variety of 
important Federal agency missions, including national security; climate 
modeling and weather prediction; modeling and simulation in biology, 
chemistry, materials science, nanoscale science and technology, and 
physics; and others. Over the years, the program evolved in scope into 
one that covers information technologies more broadly, including not 
only high-performance computing and advanced networking, but also cyber 
security and information assurance, human computer interaction and 
information management, software design, high confidence software and 
systems, and other important areas. Through this evolution, high-
performance computing not only remains the dominant element of the 
NITRD Program, but has been cited on a recurring basis as a high 
priority within the Federal R&D portfolio.
    The Office of Management and Budget (OMB) and the Office of Science 
and Technology Policy (OSTP) annually issue a joint memorandum on the 
Administration's R&D budget priorities. In the past 4 years, high-end 
computing has been identified as one of those priorities. These 
memoranda set the stage for significant focused interagency 
coordination by Federal agencies, which I will discuss further shortly, 
from the establishment in 2003 of the High-End Computing Revitalization 
Task Force that led to the development of the Federal Plan for High-End 
Computing in 2004, to directing agencies to ``aggressively focus on 
supercomputing capability, capacity and accessibility issues,'' in 
accordance with that plan.
    The Administration's support has led to significant investments in 
high-performance computing. In 2002, the funding for high-performance 
computing in the NITRD Program was less than $0.8 billion. In 5 years, 
that budget grew by over 65 percent to a Fiscal Year 2007 budget 
request of over $1.3 billion for high-performance computing R&D, R&D 
infrastructure, and applications. The National Science Foundation, the 
Department of Defense, and the Department of Energy together account 
for over $1 billion of that investment (see Table 1). In the 
Administration's FY 2007 budget, high-performance computing accounts 
for over 40 percent of the $3.1 billion NITRD Program budget request, 
and accounts for more than half of the increase in the NITRD Program 
budget from the previous year.
    The President's emphasis on science and technology, which is in 
part embodied in the American Competitiveness Initiative (ACI), is 
further contributing to the development of world-leading high-end 
computing capability and capacity, which is identified as a key goal 
for ACI research. The three agencies that are part of the American 
Competitiveness Initiative--the National Science Foundation (NSF), the 
National Institute of Standards and Technology (NIST), and the 
Department of Energy's (DOE's) Office of Science--are all members of 
the NITRD Program, and all fund high-performance computing investments.

 Table 1: Largest Government Funders of High-Performance Computing R&D,
                  R&D Infrastructure, and Applications
------------------------------------------------------------------------
                                                          FY 2007 Budget
                      NITRD Agency                         Request ($M)
------------------------------------------------------------------------
NSF                                                                  337
DOD:                                                                 375
    OSD and DOD Service organizations                                195
    DARPA                                                            118
    NSA                                                               62
DOE:                                                                 329
    Office of Science                                                296
    NNSA                                                              33
------------------------------------------------------------------------

    As a result of the ACI, the high-performance computing budget at 
NSF is expected to increase by more than $53 million above its FY 2006 
level, enabling NSF to pursue the goal of a petascale computing 
environment and resources by 2010. Similar investments at DOE's Office 
of Science are expected to increase by more than $82 million above 
their FY 2006 levels due to the ACI, which will make possible upgrades 
and diversification of existing high-performance computing platforms 
and the acquisition of a next-generation platform, at various DOE 
National Laboratories. NIST's investments are supporting the 
development of high-performance computing tools, standards, and 
algorithms, as well as research on quantum computing and secure quantum 
communications. The Defense Advanced Research Projects Agency (DARPA), 
though not part of the ACI, is another key supporter of high-
performance computing R&D, an area in which its budget is increasing by 
over $23 million above FY 2006 levels.\1\
---------------------------------------------------------------------------
    \1\ Additional detail about high-performance computing budgets, 
technical activities, and coordination activities can be found in the 
FY 2007 Supplement to the President's Budget for the Networking and 
Information Technology Research and Development Program (http://
www.nitrd.gov/pubs/2007supplement/).
---------------------------------------------------------------------------
    High-performance computing has been and continues to be a funding 
priority within the Federal R&D portfolio. Together, the guidance, 
leadership, and past and future investments in high-performance 
computing have demonstrated and solidified the Administration's 
commitment to U.S. leadership in this area.

Impact and Success of Interagency Coordination of High-Performance 
        Computing
    I would now like to take the opportunity to highlight some of the 
success stories that have emerged from the interagency coordination 
activities of the government's high-performance computing research 
community.
    Until 2003, interagency coordination of high-performance computing 
activities took place through the NITRD Program's High-End Computing 
Coordinating Group. It was then that a decision was made within the 
Administration to increase the government's focus on high-performance 
computing. In April 2003, Dr. John H. Marburger III, Science Advisor to 
the President and Director of OSTP, established the High-End Computing 
Revitalization Task Force (HECRTF) and charged this group to develop a 
Federal plan that covered high-performance computing R&D capability, 
capacity, and accessibility of high-performance computing resources; 
and procurement issues. The release of the Federal Plan for High-End 
Computing in May 2004, and the increase in visibility through elevating 
the High-End Computing Coordinating Group to an Interagency Working 
Group under the umbrella of the National Science and Technology 
Council, represented the start of a renewed emphasis on high-
performance computing within the NITRD Program.
    This cooperation, along with strong leadership from the OSTP and 
OMB, has resulted in unprecedented coordination on high-performance 
computing issues among Federal agencies. A few examples follow:

   DARPA High Productivity Computing Systems (HPCS) Program: 
        DARPA's HPCS program was established in order to develop a new 
        generation of economically-viable high productivity computing 
        systems for national security and industrial user communities 
        by the end of this decade, producing substantial advances in 
        the performance, programmability, portability, and robustness 
        of these systems. Although initiated by DARPA, this program has 
        garnered the support of over a half dozen Federal agencies 
        which have contributed to HPCS technical planning and 
        coordination, and also of the broader multi-agency research 
        community.

         More importantly, as a result of recognition that this program 
        is the government's primary effort directed at next-generation 
        high-performance computing architectures, several of these 
        Federal agencies have contributed their own funding to the 
        program, thereby increasing the leverage of DARPA's 
        investments. The HPCS program is close to entering its third 
        phase, which is aimed at development and prototype 
        demonstration. It is expected that the additional funding 
        provided by other agencies will make it possible to fund more 
        projects in Phase III than would have been possible with DARPA 
        funding alone. This will increase the diversity of 
        architectures that will be explored through this program, 
        thereby expanding the pool of concepts available on which to 
        build next-generation systems in the future, and helping to 
        cement U.S. leadership in this critical technology area.

   High-End Computing University Research Activity (HEC-URA): 
        HEC-URA is a program for funding university research that has 
        been supported by interagency planning. A group of NITRD 
        agencies has been collaborating since 2004 to identify research 
        needs for high-performance computing, and to develop programs 
        to meet those needs. Most recently, following a pair of 
        workshops held last year, a solicitation was released by NSF 
        this year to fund university research in file systems and 
        storage technologies for high-performance computing systems. 
        Though led by NSF, three other Federal agencies contributed 
        funding to support HEC-URA file systems and storage projects 
        that have direct relevance to their agency missions, helping to 
        ensure the availability of research results that would not 
        necessarily have emerged from their own agencies' research 
        programs.

   High-performance computing benchmarks, performance metrics, 
        and performance modeling: The use of benchmarks, performance 
        metrics, and performance modeling are key to a variety of high-
        performance computing issues, ranging from guiding decisions on 
        which architectures to invest in at research stages, supporting 
        procurement decisions by providing consistent bases for 
        comparing alternative systems, and predicting the performance 
        of various types of systems on different classes of computing 
        applications. Because of the importance of these issues and 
        their broad relevance to needs that are shared by multiple 
        agencies, over a half dozen Federal agencies have been 
        collaborating on the development of performance metrics, 
        measurement tools, and benchmarks, with several of these 
        agencies providing funding to support related research.

    In my preceding discussion, I have highlighted several examples of 
high-impact results of interagency coordination. These are just a few 
of the many instances of the cooperation that is taking place across 
Federal agencies and the positive effects that these collaborative 
efforts have produced. Numerous other examples are identified in the FY 
2007 Supplement to the President's Budget for the NITRD Program, which 
I referred to earlier. I would now like to close my remarks with a 
brief discussion of U.S. leadership in high-performance computing 
technologies.

U.S. Leadership in High-Performance Computing in the Context of Global 
        Competitiveness
    I described earlier the establishment of the High End Computing 
Revitalization Task Force that led to the development of the Federal 
Plan for High End Computing. Agencies are now working together to 
implement that plan, focusing on R&D programs for hardware, software, 
and systems, the different technical elements of the roadmap laid out 
in the plan. Distinctions between different classes of machines 
(capability machines, also referred to as leadership class machines, 
versus capacity machines intended to provide the high-performance 
computing capacity needed to meet government agency needs), and 
collaborative funding of programmatic activities such as those I 
described earlier, have helped make better use of Federal R&D 
investments in high-performance computing.
    The focus of the government research community on issues that 
extend beyond technical program planning is as noteworthy as the level 
of collaboration on R&D that I have described previously. In the area 
of benchmarking and performance metrics that I discussed earlier, 
agency sharing of technical results and best practices is already 
productively influencing the procurement of high-performance systems. 
The issue of accessibility of high-performance computing resources as a 
new Administration priority represents another important evolution in 
thinking within the government research community outside the direct 
scope of R&D investment. This issue emerged with the realization that 
the use of government high-performance computing resources should not 
be restricted only to the community of researchers directly funded by a 
given agency. With support from OSTP and OMB, agencies are now working 
to ensure that the use of computing resources they fund can also be 
used meet the needs of broader constituencies.
    Two notable examples of the impact of this policy change are the 
DOE's Innovative and Novel Computational Impact on Theory and 
Experiment (INCITE) program, and NASA's National Leadership Computing 
System (NLCS). Both of these agencies opened up the use of their 
systems to users outside of their traditional user community, while 
still maintaining the high standards of merit-based peer review. As a 
result, DOE awarded millions of processor hours of supercomputing time 
to four industry research projects in the latest INCITE program cycle, 
and NASA awarded a million hours on a NASA supercomputer to the 
National Institute of Standards and Technology (NIST), an agency that 
has important problems that require high-performance computing to 
solve, but which does not procure its own dedicated high-performance 
computing systems.
    In addition to making high-performance computing resources 
available to support private sector R&D, the government is working more 
generally to foster the use of high-performance computing in the 
private sector. Several government agencies are providing funding for 
the Council on Competitiveness's High-Performance Computing Initiative. 
This initiative, undertaken by this well-known nonpartisan and 
nonprofit organization, is funding studies, conferences, and 
educational activities to stimulate and facilitate wider usage of high-
performance computing across the private sector, in order to propel 
productivity, innovation, and competitiveness.
    In March 2002, issues of innovation and competitiveness in the 
context of high-performance computing gained high visibility when Japan 
brought online a new supercomputer called the Earth Simulator, which 
became at that time the world's fastest supercomputer. Although people 
in some policy circles were caught by surprise by this development, the 
system had been publicly announced long in advance and its existence 
was known by experts in the research community. Many in the research 
community called for a tempered reaction, arguing that the leap-
frogging by a Japanese supercomputer to the position of world's fastest 
machine was simply a result of the natural march of progress.
    Three weeks ago, a new version of the Top500 Supercomputer Sites 
list was released. \2\ This list, which surveys the world's 500 fastest 
supercomputers (excluding classified systems) as ranked by a well-known 
benchmark clearly confirms that the United States continues to hold a 
strong leadership position in the world of high-performance computing 
technologies. Some interesting statistics drawn from the latest Top500 
list:
---------------------------------------------------------------------------
    \2\ See http://www.top500.org/.

   The Earth Simulator, which held the number one position 4 
        years ago, now sits in the number ten position. Six of the nine 
        machines above it are in the United States, including all of 
---------------------------------------------------------------------------
        the top four machines.

   The U.S. dominates the list as a whole; 60 percent of the 
        world's 500 fastest supercomputers are installed in the United 
        States.

   U.S. vendors are the dominant suppliers of supercomputing 
        systems in the world. The top three vendors of systems on the 
        Top500 list are all U.S. companies, and account for nearly 75 
        percent of the systems on the list, including those outside the 
        U.S.

   Even foreign systems rely overwhelmingly on U.S. 
        technologies. Of the top 20 non-U.S.-based systems, 15 were 
        sold by U.S. companies. Of the remaining five that were built 
        by foreign companies, a majority were built using high-
        performance microprocessors supplied by U.S. companies.

    Looking back, we can now confidently say that while the clamor 
surrounding the launch of the Earth Simulator 4 years ago brought to 
the attention of policymakers the importance of supercomputing, it did 
not represent a pivotal crisis to U.S. global competitiveness. This is 
important to note in the context of recent announcements from Japan 
regarding an undertaking to develop a successor to the Earth Simulator, 
which will take place in phases over the next few years.

Conclusion
    The fact that the U.S. currently holds the title of world's fastest 
supercomputer does not herald a new era in U.S. leadership in high-
performance computing any more than the loss of the number one position 
implied a loss of leadership. High-performance computing has been--and 
will continue to be--a cornerstone in the government's networking and 
information technology R&D portfolio.
    The clearest demonstration of progress over the past 4 years, 
however, should not be viewed in terms of the raw speed of the world's 
fastest machine, but rather in the context of the growing focus on 
domestic high-performance computing policy, the unprecedented 
interagency coordination and collaboration on technical planning and 
implementation taking place within the government research community, 
and the increasingly cooperative ties between the government research 
community and the private sector. These latter attributes are not 
simply due to the march of technological progress, but are the result 
of focused efforts aimed at policy development, budget and technical 
planning, and the fostering of a vibrant government research community 
consisting of dedicated individuals with shared priorities committed to 
working toward common objectives.
    Once again, I thank you for the opportunity to be here today and 
would be happy to answer any questions.

    Senator Ensign. Thank you, doctor.
    Let me just start with this question, because you have 
talked about the United States and its commitment to high-
performance computing. When we compare ourselves with other 
countries, you said when Japan came online with the fastest 
high-performance computer and now we have the fastest computers 
again--at any one point in time it is probably not that 
important. What seems to be more important is the commitment to 
high-performance computing. Which country has the commitment to 
high-performance computing? How do we compare with other 
countries, other major industrialized countries? How do we 
compare to Europe, because it is hard to just pick one country 
there with the EU? How do we compare with China, EU, India, 
Japan, Korea, those types of countries?
    Dr. Szykman. Globally, I would say that Japan is probably 
at a leadership position for making a national commitment to 
high-performance computing. They recently, earlier this year 
announced a national technology area of importance in the 
context of high-performance computing. That is a high-level 
policy commitment. I would say that that commitment is shared 
within the United States.
    The Office of Science and Technology Policy and the Office 
of Management and Budget annually issue a memo, a guidance memo 
to agencies, that directs agencies to focus on certain R&D 
priorities, and high-performance computing has been identified 
as a priority in the memo in each of the past 4 years. So there 
is certainly very high-level White House commitment to high-
performance computing in this country.
    Senator Ensign. How does that translate down into other 
countries dollarwise versus our country dollarwise, investment 
by the government?
    Dr. Szykman. I would probably have to check some figures 
and I would be happy to get back to you.
    Senator Ensign. Could you get that for us?
    Dr. Szykman. I can say that in terms of countries in Asia 
besides Japan, there certainly is an interest in high-
performance computing, but not a local, national capability for 
developing very high-end systems in, for example, Japan--I am 
sorry, in for example China or India.
    Senator Ensign. Thank you.
    You mentioned performance metrics and I have always been 
interested in performance metrics throughout our government. I 
think it is very important that when we are putting money into 
something we try to at least measure what we are getting for 
our investment as much as possible. It is not always possible, 
but we should make every attempt to measure that.
    In your testimony you talk about a 65 percent increase 
since 2002 for funding for high-performance computing. Do you 
have any metrics that explain the increase? That is a fairly 
large increase, 65 percent in a very short period of time. Do 
you have any metrics? You talk about metrics. Now do you have 
any metrics to show us what we are getting back as a Nation for 
that investment? Sorry to use simple terms, but hopefully you 
understand the essence of the question.
    Dr. Szykman. Certainly. The work that is being done 
collaboratively in the area of metrics has in fact been aimed 
at being able to make better decisionmaking about investments 
in the area of high-performance computing. Developing better 
predictions of different alternative architectures helps people 
decide what areas to invest in, both at the research level as 
well as informing procurement decisions for more advanced 
systems.
    I could probably do some research and come up with 
particular examples, but certainly if we look at the fastest 
machine that is online today funded by the U.S. Government, it 
is considerably more advanced than the one that was in place in 
the past, and the performance metrics and benchmarks that are 
being developed today are being incorporated in calls for 
proposals as well as procurement issues for future systems.
    Right now the most advanced program in the area of 
developing next-generation technologies will be making use of 
significantly improved benchmarks over what was available just 
a few years ago.
    Senator Ensign. A lot of this information on high-
performance computing is new to me. So let us just take even 
the top five or top ten computers in the world today as far as 
performance capabilities, can you give the Subcommittee an idea 
what it costs to put one of those computers together? I realize 
there is a lot of R&D, but from start to finish what does it 
cost? If the computer is going to be put in a university 
someplace, what kind of investment are we looking at for one of 
these, especially the very, very fast ones, and maybe also for 
computers that are within the second hundred in terms of speed 
and processing ability, how the costs would compare between 
these two types of computers?
    Dr. Szykman. I would say that that would probably be an 
ideal question for some of the industry members of the second 
panel. They are the ones who sell these machines and could give 
a more informed answer than I could.
    In terms of the R&D that leads up to the development of 
these systems, though, it is on the order of hundreds of 
millions of dollars of R&D investment on the private sector 
side to support next-generation architecture development.
    Senator Ensign. Senator Cantwell.
    Senator Cantwell. Thank you, Mr. Chairman.
    Mr. Szykman, I have a couple questions. First of all, did 
you support the legislation that moved through the House of 
Representatives? Did the Administration support that language?
    Dr. Szykman. Yes, the Administration was supportive. You 
are referring to H.R. 28?
    Senator Cantwell. Yes.
    Dr. Szykman. Yes. And we did provide a couple of comments 
and suggestions, but in general we did support that 
legislation.
    Senator Cantwell. Why do you think that it is important to 
upgrade the coordination? What specifically do you think that 
that legislation gets at?
    Dr. Szykman. I would say one of the benefits of that 
legislation is to revise the statutory descriptions of the 
NITRD Program. If we go back 15 years to when the program was 
initially established, it was focused on high-performance 
computing and the advanced networking to support those high-
performance computing centers. The program over the past 15 
years has broadened considerably into new program component 
areas and certainly having an expansion of the scope of the 
program in legislation was helpful, as well as the 
rearticulation of the priority of some of these technology 
areas.
    Senator Cantwell. What do you think those priorities are in 
the technology areas?
    Dr. Szykman. High-performance computing, as I mentioned in 
my testimony, remains one of the main priorities and in fact 
accounts for 40 percent of the program budget, even though 
there are eight different program component areas in the 
program. Other R&D priorities in the area of IT R&D include 
advanced networking, large-scale networking, which is clearly 
needed to support interconnectivity between high-performance 
computing centers as well as more general research facilities 
and the connectivity needed for users to access those 
facilities.
    Senator Cantwell. What areas do you think the United States 
right now has a lead in in the area of high-performance 
computing research?
    Dr. Szykman. I would say the United States has a strong 
lead in most or all areas. High-performance computing is not an 
area that really can be looked at as a collection of individual 
areas. It needs to be looked at holistically at the level of 
architectures and systems, and the most advanced systems are 
the ones that are being developed through R&D in the United 
States. Those systems include aspects of hardware, aspects of 
software, aspects of storage systems, as well as the overall 
architectures needed to bring these together into functioning 
systems.
    Senator Cantwell. So there is not an area of concern that 
you have where the United States may be losing an R&D advantage 
because dollars are going to another country because of their 
particular focus in an area of supercomputing?
    Dr. Szykman. I would say through interagency coordination 
the government research community is able to identify weak 
areas in the overall portfolio and put funding in those areas. 
In fact, one of the areas that was not highlighted as a high 
priority in the ``Federal Plan for High-End Computing'' was the 
area of storage systems for high-performance computing. That 
was mentioned in there, but was not highlighted as a strong 
priority. However, over the couple of years since the release 
of that plan interagency coordination has identified that as an 
important need. Agencies have come together to fund university 
research in that area with funding from multiple agencies to 
help fill in some of the gaps.
    So I would say in summary that the interagency coordination 
mechanisms that are in place are very effective at identifying 
needs for the future of high-performance computing R&D and are 
able to direct funding and technical planning against those 
needs.
    Senator Cantwell. So you do not have a particular area? For 
us in Washington State, since we see so much that goes on with 
computing in general and supercomputing and systems biology, I 
would say that we are doing pretty well there as a country and 
having an advantage in driving R&D investment. But there are 
probably some other areas that I would say I am not so sure. I 
mean, the Japanese have taken some lead on various models as it 
relates to weather and climate; is that not correct?
    Dr. Szykman. In the area of applications, I would say that 
most likely agencies themselves could provide clearer 
information than I can from my office. We are focused more 
typically on the R&D and less on the particular applications 
that are being done, even though we collect funding information 
on those applications. But certainly the most advanced high-
performance computing models that are in place today are being 
supported by high-end capabilities funded by the U.S. 
Government.
    In the area of climate modeling and weather prediction, for 
example, NOAA, the National Oceanographic and Atmospheric 
Administration, has recently over the past few years very 
significantly upgraded its capabilities for doing modeling and 
prediction in ways that allow longer term modeling of climate 
and weather, and these are things that are having a direct 
influence on people every day when they turn on the weather in 
the morning.
    So the U.S. is, I believe, maintaining its leadership in 
the applications areas as well.
    Senator Cantwell. I know you mentioned advanced networking. 
Do you believe that there is an under-investment in software 
for high-performance computing?
    Dr. Szykman. That is one of the other areas that I think 
that agencies recognized as being an important area that in the 
past, if we look back perhaps 3, 4 years, had been somewhat 
underfunded, and there is a renewed interest in putting funding 
in those areas and software programs within different agencies, 
including the National Science Foundation and the Department of 
Energy.
    Senator Cantwell. Mr. Chairman, since I am on a time limit 
and I have two witnesses I would like to hear from, I could ask 
Mr. Szykman more questions, but I think I would just file those 
for the record and thank him for his testimony and allow us to 
hear from some of the other individuals that are here with us 
today.
    Senator Ensign. I would like to thank you for your 
testimony. We will have other questions for the record, and we 
appreciate your being here.
    Let us call the second panel to the table. I am just going 
to introduce all six of you at once and we will go in the order 
of your testimony. All of you can please come up now, and I 
will introduce all of you at the same time. Then we will hear 
from you in the order in which I introduce you.
    Our first witness on this panel will be Dr. Irving--and 
this is going to be a tough name----
    Dr. Wladawsky-Berger. ``Vla-DOW-skie.''
    Senator Ensign. ``Vla-DOW-skie''-Berger. How is that?
    Dr. Wladawsky-Berger. Perfect.
    Senator Ensign. He is IBM's Vice President for Technical 
Strategy and Innovation.
    After Dr. Wladawsky-Berger will be Mr. Christopher Jehn. He 
is the Vice President of Government Programs at Cray 
Incorporated. The next witness after that will be Mr. Jack 
Waters. Mr. Waters is the Executive Vice President and Chief 
Technology Officer of Level (3) Communications.
    After that, we will hear from Joseph Lombardo. Mr. Lombardo 
is the Director of the National Supercomputing Center for 
Energy and the Environment at the University of Nevada, Las 
Vegas, my alma mater.
    Our next witness after that will be Mr. Michael Garrett, 
who is the Director of Airplane Performance, Boeing Commercial 
Airplanes, for The Boeing Company. And our final witness will 
be Dr. Stanley Burt. Dr. Burt is the Director of the Advanced 
Biomedical Computing Center in Frederick, Maryland.
    So we can make sure to get in all the questions that we 
possibly can, because we have all of your written testimonies, 
if you could sum up your testimonies in about 5 minutes, that 
would give us a chance to hear from each one of you and then 
allow plenty of time for further discussion. So I appreciate 
each one of you being here and look forward to your testimony. 
Let us start with Doctor Wladawsky-Berger.

           STATEMENT OF DR. IRVING WLADAWSKY-BERGER,

       VICE PRESIDENT, TECHNICAL STRATEGY AND INNOVATION,

       INTERNATIONAL BUSINESS MACHINES CORPORATION (IBM)

    Dr. Wladawsky-Berger. Thank you. On behalf of IBM I want to 
thank you, Mr. Chairman, for the opportunity to testify. I am 
Vice President for Technical Strategy and Innovation at IBM and 
have been involved with supercomputing initiatives for over 20 
years. With your permission, I will simply summarize my written 
testimony.
    Today IBM leads the industry with the world's top three 
supercomputers and almost half of the world's top 500 
supercomputers. We were first to deliver over 100 teraflops, or 
100 trillion operations per second, 1012 of peak 
performance, to the DOE's Lawrence Livermore National Lab, 
where we also first demonstrated the practicality of using well 
over 100,000 microprocessors on a single problem.
    Likewise, we have been working with DARPA to help them make 
very high-end systems more productive, and are investing in 
advanced hardware and software that will culminate in a system 
capable of more than one sustained petaflop, 1015.
    In the process of all this, we have learned many lessons, 
but two are especially significant. First, it is vital to work 
closely with lead partners in research labs and universities to 
push the envelope of performance, applications and discovery. I 
can not overemphasize enough from my personal experience the 
importance of this pilot to developing working systems for real 
research on important applications, and the fact that it is the 
Federal Government that is instrumental in creating them.
    Second, the marketplace is all-important. Many 
supercomputing companies have failed because they relied solely 
on government-based projects and were heedless of marketplace 
requirements that go beyond leading performance to competitive 
prices, energy efficiency, and sophisticated software and 
applications. While we are very proud of IBM's leadership in 
supercomputers, we are equally proud that it is an actual, 
viable business for us with clients around the world.
    Why is a national supercomputing capability vital to the 
U.S.? Supercomputing systems and applications push the envelope 
in multiple dimensions. They analyze huge amounts of 
information. They accurately simulate both the natural world 
and the world of manmade objects, and they present the results 
in highly visual and realistic ways so we can interact with 
them.
    Additionally, supercomputing architectures and applications 
foreshadow the future. If one is removed from the advanced 
research, new ideas, and creative minds in supercomputing, one 
will inevitably misread the major trends in computing.
    Finally, supercomputers enable scientists to make 
discoveries that would be difficult, perhaps impossible, to 
accomplish experimentally.
    The supercomputing market has changed radically in the last 
decade. It was once a niche market because the hardware was so 
expensive, but that has all changed with the introduction of 
workstation and PC-based technologies. Today we are even 
working to build supercomputers with technology from the worlds 
of consumer electronics and video games. All these approaches 
use components from high volume markets, thus the costs are 
significantly lower than in the early days and the potential 
markets are much, much bigger.
    My basic point is this: A commercial business model has 
reduced costs and enabled us to address the vast spectrum of 
public- and private-sector applications. One-off machines built 
for a single mission are usually very expensive, impractical in 
the marketplace, and not viable in the long term.
    Progress in supercomputing hardware has been outstanding. 
The real challenge, however, lies in both application software 
and system software, as has been widely recognized in a number 
of studies. Software is so consequential because 
supercomputing's value is not in the technology, important as 
it is, but in its applications, which makes software critical. 
My formal testimony reviews the progress and promise of some 
key applications. There is enormous promise both in classic or 
more mature applications, such as defense and national 
security, science, engineering, and weather and climate, and in 
the newer applications that are so vital to the national 
interest, like energy, healthcare and bioinformatics, learning 
and training, and business in general.
    In civilian nuclear energy, for example, the GNEP, or 
Global Nuclear Energy Partnership, and ITER, International 
Thermonuclear Energy Research, programs are excellent examples 
of government-industry academic collaboration on matters of 
national and international importance, market relevance, and 
timeliness. They deserve support.
    Supercomputing today is essential for innovation in both 
the world of science and the world of commerce. It is an 
indispensable tool if our country is to thrive in a global 
economy that grows more competitive by the day. It is therefore 
essential to pass an innovation authorization bill this year, 
as in S. 2802, as you, Senator Ensign, Senator Stevens, and 
others on this committee know.
    The Federal Government funds basic research and establishes 
priorities for research in the pursuit of innovation and 
competitiveness. That makes wise policy choices critically 
important to a national supercomputing capability. To realize 
that capability, Congress should clearly outline and invest in 
a long-term strategy. For example, the President's budget 
request for Fiscal Year 2007 includes high-performance 
activities that range from biomedical computing to Earth and 
space science research and many others. Clear direction and 
consistent funding will foster investment by industry and 
academia, so that together we can address the challenges that 
face our country and grow the capabilities of our knowledge. 
This is the kind of joint effort between government, 
universities, and private industry for which there is no 
substitute.
    Thank you.
    [The prepared statement of Dr. Wladawsky-Berger follows:]

  Prepared Statement of Dr. Irving Wladawsky-Berger, Vice President, 
  Technical Strategy and Innovation, International Business Machines 
                           Corporation (IBM)

Introduction
    On behalf of IBM, I would like to thank the Subcommittee, and 
especially Chairman Ensign and Senator Kerry, for the opportunity to 
address the evolution of supercomputing during the 1990s, the 
priorities we should focus on, and the challenges we face.
    First, by way of introduction, I am currently Vice President for 
Technical Strategy and Innovation at the IBM Corporation, and am 
responsible for identifying emerging technologies and marketplace 
developments critical to the future of the IT industry, and organizing 
appropriate activities in and outside IBM in order to capitalize on 
them.
    My association with computers began in the Summer of 1962, when 
prior to entering the University of Chicago, I was employed at the 
computation center where I worked part-time through my college years 
doing scientific programming. I later went on to get a Ph.D. in 
Physics, and did my research on computational atomic and molecular 
physics. After finishing my Ph.D. at the University of Chicago, I 
decided that I was better suited for computing than for physics, 
switched fields and joined the Computer Sciences Department at IBM's 
Thomas J. Watson Research Center in June 1970.
    By then, IBM had decided, for a variety of reasons, to exit the 
scientific computing market, where it had been a leader for a number of 
years. We re-entered the market in the second half of the 1980s by 
adding what is called a vector feature to our mainframes, and a few 
years later in the early 1990s we became a leader in the emerging area 
of parallel supercomputing with our SP system. I was the General 
Manager of both these efforts.
    Today, according to the authoritative Top500 Supercomputing 
rankings, IBM:

   Leads with the world's top three supercomputers: BlueGene/L 
        for the U.S. Department of Energy (DOE) with 280.6 sustained 
        teraflops (trillion floating point operations per second), 
        BlueGene/W at Watson Research with 91.3 sustained teraflops (or 
        114.7 teraflops peak) and DOE's ASC Purple at Lawrence 
        Livermore National Laboratory with 75.8 sustained teraflops (or 
        92.8 teraflops peak).

   Has supplied 240 of the world's top 500 supercomputers--more 
        than any other vendor.

   Accounts for over 1.5 petaflops of aggregate performance in 
        the TOP500 list (from a total of 2.79 petaflops).

   Has supplied more supercomputing systems than any other 
        vendor in the Top10, Top20, Top100 and Top500.

   Has supplied the most cluster systems with 177 of 364 (48.6 
        percent).

   Has built the largest university supercomputer in the U.S.--
        Big Red--a Cluster rated at 15 teraflops and installed at the 
        University of Indiana.

    IBM has held the number one spot in the Top500 list since June 
2005. Japan's Earth Simulator held the number one spot for the previous 
3 years. IBM was the first to deliver a system that achieved over 100 
teraflops--that is a system that could perform over 100 trillion 
operations per second--of peak performance; in fact BlueGene/L has 
tested at 360 teraflops (peak). And we have achieved actual sustained 
performance of from 100 to 200 teraflops on a number of applications of 
real importance to the National Nuclear Security Administration. The 
BlueGene/L supercomputer has been measured at 10-times the energy 
efficiency (measured by Watts of electricity needed to attain a 
particular level of performance) of any of the top 20 supercomputers, 
and it is similarly efficient in its space requirements. In the 
process, we have demonstrated the practicality of using well over 
100,000 microprocessors and then leveraging their computational 
capability efficiently on a single problem.
    Along the way, we have learned many lessons, but I believe that two 
are especially significant. First, it is vitally important to work 
closely with lead partners in research labs and universities in order 
to ``push the envelope'' in terms of performance, applications and 
discovery. I cannot overemphasize the importance of these leading-edge 
pilots in propelling us forward and bringing together all the elements 
needed to develop working systems that can be used for real research in 
important application areas.
    The second key point is the importance of the marketplace in 
guiding our actions. Through my professional career I have seen many 
supercomputing companies fail because they relied solely on government-
based projects and were heedless of marketplace requirements for, not 
only leading-performance, but competitive prices and sophisticated 
software and applications as well. While we are very proud of IBM's 
leadership in the Top10, Top100 and Top500, we are equally proud that 
supercomputing is a viable business for us with many clients around the 
world in both the private and public sectors.

What Is Supercompting?
    Supercomputing is defined by three key characteristics. First, the 
applications are information-intensive; second, they deal with 
computation-intensive simulations--both in the natural world of 
physics, chemistry and biology, and in virtual worlds, such as 
engineering objects and entertainment; third they enable the 
visualization of information and simulations so people can interact 
with the results--as exemplified by scientific visualization and the 
more recent emergence of video games played between myriad 
participants.

Why Is Supercomputing so Important?
    Through the years, we have come to realize that supercomputing 
architectures and applications foreshadow the future of computing 
itself. Indeed, if one is removed from the advanced research, new ideas 
and creative minds in supercomputing, one will inevitably misread the 
major trends in computing. This is among the main reasons IBM re-
entered the market in the late 1980s.
    Beyond its role as a precursor, supercomputing has become essential 
to the pursuit of scientific inquiry. To quote the June 2005 report by 
the President's Information Technology Advisory Committee (PITAC), 
``Computational science has become the third pillar of the scientific 
enterprise, a peer alongside theory and physical experiment.'' 
Supercomputers enable scientists to either make discoveries that would 
be difficult (perhaps impossible) to accomplish experimentally or to 
point researchers in new directions.
    Examples abound. They include (but are certainly not limited to) 
developing insight into the behavior of materials under extreme 
conditions that cannot be reproduced experimentally, enabling 
scientists to make reliable predictions about the behavior of our 
nuclear stockpile or the safety of aging nuclear reactors, for example. 
Supercomputers can also find previously undiscovered sequences in so-
called ``junk DNA'' that may lead to new insights into its 
``function.'' They can also discover ``docking sites'' for new drugs, 
i.e., receptors on molecules where a drug can potentially attack a 
disease. Or supercomputers can perform multi-century simulations to 
understand trends in the Earth's climate.

Growth of the Supercomputing Market
    Supercomputing was once confined to a niche market, because the 
hardware was so very expensive. That changed over time with the 
introduction of workstation and PC-based technologies, the latter 
becoming immensely popular in Linux clusters during the late 1990s. 
Today, we even use low-power, low-cost micros--consumer-based 
technologies--to attain very high degrees of parallelism and 
performance, as in our Blue Gene system, which has reached a peak of 
360 trillion calculations per second. Now, we are seeking to build 
supercomputers using technologies from the gaming world, such as the 
Cell processor.
    All these approaches leverage components from high-volume markets, 
and aggregate them using specialized architectures; thus the costs are 
significantly lower than in earlier days and the potential markets are 
consequently much bigger.
    Progress in supercomputing hardware has been nothing short of 
astounding. The real challenge, however, is software, both application 
software and systems software. In fact, both the 1999 PITAC report, 
with which I was personally involved, and the June 2005 PITAC report 
made precisely that point.

Key Application Areas
    But the real value of supercomputing to society is not in the 
technology, architecture and software, important as they are. The value 
of supercomputing is best appreciated by considering its application, 
so let me review the recent progress and the promise in a few key 
application areas, starting with the ``classic'' or more mature ones 
and then moving on to some of the newer opportunities.

Defense and National Security
    Let me start by discussing Blue Gene/L and ASC Purple, two of the 
world's top three supercomputers, residing at Lawrence Livermore 
National Laboratory (LLNL). They are vital to the National Nuclear 
Security Administration's (NNSA) Advanced Simulation and Computing 
Program (ASC), which in turn is an essential element of our Nation's 
Stockpile Stewardship Program. ASC provides the integrating simulation 
and modeling capabilities and technologies needed to combine new and 
old experimental data, past nuclear test data, and past design and 
engineering experience into a powerful tool for future design 
assessment and certification of nuclear weapons and their components.
    Already, the simulation and modeling tools are improving the 
assessment of stockpiles far in advance of schedule. Indeed, weapons 
designers, scientists, and engineers now rely on ASC simulation and 
modeling capabilities and technologies to assess changes occurring in 
aging stockpiles of nuclear weapons and to assess and certify planned 
refurbishments of weapons system components.
    On March 9, 2006, Lawrence Livermore National Laboratory and IBM 
announced a fundamental breakthrough using ASC Purple. IBM and LLNL 
demonstrated over 102 gigabytes per second of sustained read and write 
performance to a single file using specialized software that 
orchestrates thousands of processors and thousands of disk storage 
devices. The breakthrough is expected to stimulate development of data-
intensive applications in areas like customized medicine, online 
gaming, entertainment, and homeland security, as well as in traditional 
high-performance computing applications.
    Then on June 22 of this year, the NNSA announced that it had 
achieved an unprecedented level of performance using our Blue Gene/L. 
This world record for a scientific application was set by achieving a 
sustained performance of 207.3 teraflops, running ``Qbox'' computer 
code for conducting materials science simulations critical to national 
security.

Science
    In addition, the unmatched cost-effective computational capability 
of Blue Gene has already resulted in new insights in biology.
    The scientists at the T.J. Watson Research Center have applied 
supercomputing to demonstrate that ``junk DNA'' could have very 
startling ramifications on cell regulation and species evolution. In 
another computational experiment, they have shown that a single 
mutation in a protein can render it unstable, causing it to misfold. 
Similar techniques and computational models can be applied to better 
understand fatal diseases.
    ASTRON, in the Netherlands, is using the Blue Gene supercomputer to 
develop a new type of radio telescope capable of looking back billions 
of years. This research project will enable scientists to examine the 
beginnings of the earliest stars and galaxies after the formation of 
the universe in the wake of ``the Big Bang.''
    Blue Gene/L will give ASTRON the flexibility and unparalleled speed 
it needs to gather and analyze information from its Low Frequency Array 
(LOFAR) ``software telescope'' network. Unlike current observatories 
that use big optical mirrors or radio dishes to point to distant 
galaxies, ASTRON will harness more than 10,000 simple radio antennas 
spread across the northern Netherlands and into the German state of 
Lower Saxony and interpret them using high-speed calculations.
    In many domains, theory, experimental capabilities, and 
computational advances are coming together in a manner that will 
significantly accelerate scientific discovery.

Weather, Climate
    Supercomputing is also taking weather forecasting, modeling and 
research to new levels. Research groups at several government agencies 
and research laboratories are moving traditional models to scalable 
supercomputer systems. These models are then used to test the validity 
of our current understanding of the physics of weather and to develop 
more detailed, robust, high-resolution models. When the models are 
considered trustworthy, they are used for operational forecasting by 
the National Weather Service and by environmental analysts to assess 
air quality.
    In addition, there is emerging a generation of localized, high-
resolution weather prediction capabilities customized for application 
one to 2 days ahead of time by businesses with weather-sensitive 
operations. Industries that would benefit are as diverse as aviation, 
agriculture, broadcasting, communications, energy, insurance, sports, 
entertainment, tourism, construction and others in which weather is a 
crucial factor. Extremely fast, ultra precise weather forecasts would 
be invaluable to these businesses' day-to-day decisionmaking. Such 
forecasts could be used for competitive advantage or to improve 
operational efficiency and safety.

Engineering
    Automobile companies run virtual car crashes using complex 
supercomputer simulations to ascertain how different designs react in 
collisions. This reduces the number of costly prototypes, and speeds 
the delivery of new models. With new regulations on safety in the auto 
industry and buyer preferences for safer cars, keeping this competitive 
advantage is of paramount importance to manufacturers.
    Supercomputing is also being used to create more fuel-efficient 
automobile designs. Exa Corporation, a global provider of wind tunnel 
design simulation software uses our supercomputers to help major 
automotive manufacturers and smaller suppliers solve larger, more 
complex aerodynamic, acoustic and thermal engineering problems. With 
virtually unlimited amounts of compute capacity available as needed, 
Exa's clients can perform more analysis in less time--improving quality 
and time-to-market and overall competitiveness.
    Seismic imaging is an application critical to our energy future. 
Seismic imaging is the process by which acoustic waves are generated 
and their reflections off the Earth's subsurface are collected. Seismic 
imaging applications then convert the reflected waves into a 3D image 
of the subsurface, revealing an image of a petroleum reservoir. This 
process is used by all major oil and exploration companies. Good 
quality seismic imaging is critical since dry holes can cost millions--
in the deep waters of the Gulf of Mexico as much as $100 million.
    IBM and Compagnie Generale de Geophysique (CGG), a world leader in 
geophysical services, recently announced deployment of Europe's most 
powerful seismic supercomputer to respond to growing global demand in 
the petroleum industry. The system is expected to significantly reduce 
processing times from the moment the geophysical data is collected to 
the point when it generates a seismic image. This clustered 
supercomputer will also allow CGG to boost its worldwide computing 
capacity to a maximum of 113 teraflops, and give them an unprecedented 
ability to respond to the extremely high performance requirements of 
the oil industry.
    This supercomputer installation is a result of CGG's need to 
continually improve its performance in response to the demands of a 
highly competitive market by optimizing the quality and speed of 
processing in specific applications. The new system is being deployed 
at the company's premises in Massy (France), London, Kuala Lumpur and 
Houston, Texas.
    Let me now focus on some of the newer application opportunities.

Energy
    GNEP (The Global Nuclear Energy Partnership) is a Presidential 
initiative to establish nuclear energy as the preferred emissions-free 
alternative source of electric power. By reprocessing spent nuclear 
fuel and recycling it for reuse in nuclear power plants we can control 
the process and share recycled fuel and technology with developing 
countries that need inexpensive energy. The United States has been 
encouraged in this effort by China, France, Japan, Russia, and the 
United Kingdom as well as the International Atomic Energy Agency 
(IAEA).
    Computer simulation will be essential to the success of GNEP, 
allowing us to rapidly test innovative approaches and improve our 
ability to understand and control very sensitive materials. The 
President has requested $250M in FY07 for the program, while the House 
has recommended $150M and the Senate $250M. We support the effort and 
the concept of ``making nuclear energy a renewable source of power''.
    ITER is a large international fusion experiment aimed at 
demonstrating the scientific and technological feasibility of fusion 
energy and at trying to answer the question: Can we produce practical 
amounts of fusion power on Earth? In fusion, heavy forms of hydrogen 
are fused at high temperatures with an accompanying production of heat 
energy. ITER is a step beyond the study of plasma physics and toward 
the possibility of fusion power plants actually producing electricity 
and hydrogen.
    The international project is made up seven partners including the 
United States, China, the European Union, India, Japan, Russia and 
South Korea. The facility will be housed at a site in Cadarache, 
France. We support the United States' participation in ITER and the 
funding requested by the President in the Fiscal Year 2007 budget.
Bioinformatics and Computational Biology
    These involve the use of techniques from applied mathematics, 
informatics, statistics, and computer science to solve biological 
problems. Genomes (an organism's complete information set) are 
sequenced and assembled, and then become candidates for data mining. 
This data mining is often referred to as bioinformatics.
    The objective is a better understanding of the relationship between 
specific genes and diseases, an understanding that is essential to the 
development of therapies. The point is to develop drugs that will 
target specific genes and focus on a specific disease. With the volume 
of genetic data proliferating, it long ago became impractical to 
analyze DNA sequences manually. Today, computer programs search the 
genome of thousands of organisms, containing billions of nucleotides.
    Bioinformatics has great potential for expediting delivery of new, 
individualized therapies to patients.
    Brain research is another promising scientific pursuit utterly 
dependent on supercomputing. It is also the purpose of a joint research 
initiative between the Ecole Polytechnique Federale de Lausanne (EPFL) 
and IBM. Nicknamed the ``Blue Brain Project,'' it is intended to take 
brain research to a new level.
    Scientists from EPFL and IBM are working together using the huge 
computational capacity of Blue Gene to create a detailed model of the 
circuitry in the neocortex--the largest and most complex part of the 
human brain. By expanding the project to model other areas of the 
brain, scientists hope eventually to build an accurate, computer-based 
model of the entire brain.
    Relatively little is actually known about how the brain works. 
Using the digital model, scientists will run computer-based simulations 
of the brain at the molecular level, shedding light on internal 
processes such as thought, perception and memory. Scientists also hope 
to understand more about how and why certain microcircuits in the brain 
malfunction--a failure thought to be the cause of psychiatric disorders 
such as autism, schizophrenia and depression.

Health Care
    Medical science increasingly relies on advanced information systems 
to share information, mine that information for trends and insights, 
and use those findings to head off disease or improve treatment. This 
takes sophisticated computer hardware and software, and the technology 
has advanced to the stage where truly wondrous things that yesterday 
were only wishful thinking can now be tackled.
    For example, The Scripps Research Institute and IBM researchers are 
working on new technology to anticipate, manage and contain infectious 
diseases like avian flu. Using Blue Gene, they are trying to devise a 
way to track the emergence of new virus strains and map human and 
animal responses to them. This capability will help scientists and 
governments to better understand viruses and respond effectively to 
potential pandemics. It could also enable vaccines to be created 
quickly enough to prevent massive outbreaks.
    Likewise, QuantumBio Inc., a provider of software tools for drug, 
biotechnology, and pharmaceutical companies, uses the Blue Gene 
supercomputer to help satisfy its testing needs. With Blue Gene, 
QuantumBio is able to provide users with the opportunity to study 
molecules of interest over a secure and integrated system on an as-
needed or on-demand basis.

Business
    SmartOps, a leading provider of enterprise-class supply chain 
optimization solutions for the manufacturing and distribution 
industries, used the Blue Gene supercomputer to port and test their 
Multistage Inventory Planning and Optimization (MIPO) solution in 
preparation for offering a large-scale hosted solution for their 
clients.
    Likewise, in the Finance Industry the most competitive firms are 
those that can maximize returns and minimize risk, all in the shortest 
time possible. Key to success is the ability to apply computational 
power to increasingly complex and demanding business processes. 
Workloads such as risk management, portfolio analysis, derivatives 
pricing and actuarial simulations can all benefit from the application 
of supercomputing's greater computational power.

Learning
    Highly realistic, visual interfaces first appeared with scientific 
applications as well as with flight simulators used to train pilots and 
with war game simulators used to train military personnel. These visual 
interfaces (along with the accompanying sounds) have been increasingly 
enhanced with digital animation and video games. Video games are 
particularly important because in addition to their very realistic 
visual images and sound effects, they are also highly interactive and 
increasingly collaborative, and thus a good launch pad for thinking 
about how people can best interact with all kinds of computer 
applications as well as with each other in the future. Furthermore, the 
success of video games with millions of people has stimulated the 
introduction of very inexpensive and powerful technologies, such those 
around Microsoft's Xbox and Sony's upcoming Play Station 3.
    The new highly visual, realistic, and interactive interfaces now 
hold the promise of sparking a major round of innovation for computer 
applications in general, both in rethinking how to best integrate these 
new kinds of visual interfaces with existing applications, as well as 
inspiring whole new categories of applications that we cannot even 
envision today.
    One application area that holds great promise is learning across 
the broad spectrum of needs, from K-12 all the way to the introduction 
of sophisticated new procedures for professionals. After all, since our 
brains are wired for sight and sound, these new applications should be 
able to approach humans on human terms, and thus significantly 
facilitate the learning process.

Conclusion
    Clearly, supercomputing has advanced to the point of being 
essential in myriad endeavors, in the laboratory certainly but most 
assuredly in the commercial world as well. It is indispensable to the 
process of innovation and to the ability of the United States to thrive 
in a globalized economy that grows more competitive by the day--
something The National Innovation Act of 2005 (S. 2109), which we 
support, is meant to foster.
    The Federal Government has significant influence in setting the 
agenda for basic research and in turn the use of high-performance 
computing in pursuit of innovation and competitiveness. We, in our 
industry participate in that agenda as partners. In order to realize 
the full benefits for our country, Congress, in partnership with the 
industry, should clearly outline and invest in a long-term strategy. 
For example, the President's budget request for Fiscal Year 2007 
includes high-performance computing activities funded by the Networking 
and IT Research and Development (NITRD) agencies including the National 
Science Foundation, Department of Energy and the National Aeronautics 
and Space Administration. These activities range from biomedical 
computing to Earth and space science research to weather modeling 
frameworks.
    Clear direction and consistent funding will prompt industry and 
academia to invest as well, and in partnership we can address many of 
the serious challenges that face our Nation. In the process, we will 
expand and deepen our knowledge of much of the world around us and our 
ability to influence it. These kinds of efforts unite government, 
universities and private industry in a productive collaboration--a 
partnership for which there is no substitute.

    Senator Ensign. Thank you. Our next witness will be Mr. 
Jehn.

   STATEMENT OF CHRISTOPHER JEHN, VICE PRESIDENT, GOVERNMENT 
                      PROGRAMS, CRAY INC.

    Mr. Jehn. Thank you, Mr. Chairman, for inviting me here on 
behalf of Cray. And I would like to thank Senator Cantwell too 
for those generous words about Cray. I also would like to thank 
you for holding a hearing addressing this important subject.
    I have submitted a written statement for the record and 
will briefly summarize it here. I have just a very 
straightforward story that contains only four key points. 
First, supercomputing is vitally important, as both you and 
Senator Cantwell recognized. It is key to many critical 
national security missions and it is essential for the 
country's scientific leadership and our global economic 
competitiveness.
    Second, progress in supercomputing technology has slowed. 
As we have increasingly relied on commercially-available parts 
for supercomputers, we have come to realize that those 
solutions are not always the best for the most demanding 
technical and scientific applications in government, industry, 
and academia.
    Third, the Federal Government has recognized this reality. 
In a series of recent reports, the government has recognized 
that a vital supercomputer industry is important and that U.S. 
Government support is necessary to achieve that end. These 
reports all cite the need for a systematic research and 
development program that supports R&D in the supercomputer 
industry. Industry cannot do it alone because the market for 
supercomputers is simply not deep enough to justify the kind of 
investment, the amount of investment, necessary to sustain 
progress in this area.
    Fortunately, the government is doing more than just writing 
reports. The Department of Energy has recently announced its 
intention to develop and deploy a petascale computer, that is, 
one capable of performing 1,000 trillion calculations per 
second. The National Science Foundation has announced a similar 
intent.
    Meanwhile, the Defense Advanced Research Projects Agency 
and the National Security Agency are supporting R&D. For 
example, DARPA's high-performance computing systems program is 
aimed at developing a commercially-viable system by the end of 
this decade, a system that can deliver sustained petaflops of 
performance. It would be more productive than today's 
computers, but also, equally important, more robust, use power 
more efficiently, be much easier to program, and be available 
and applicable to a much wider range of applications.
    Fourth, at Cray we understand all these problems and 
believe we have developed a vision, a plan that we are now 
acting on, to develop what we call adaptive supercomputers. 
These will be supercomputers that will combine in one system 
multiple processor technologies, so that the computer can adapt 
to the scientists' requirements, rather than demanding that the 
scientist, adapt their science to the available supercomputer.
    I would like to conclude by urging the Congress to fully 
fund the current Administration initiatives in this area. I 
would also encourage the Administration to build on these 
initiatives and develop and fund an R&D program like those 
described in the reports I cited above.
    In conclusion, I would like to thank you again for holding 
this hearing, and also thank the Congress and the 
Administration for their leadership in supercomputing over the 
past several years. There has been a lot of progress and we 
need to build on that momentum. The time to invest is now.
    [The prepared statement of Mr. Jehn follows:]

        Prepared Statement of Christopher Jehn, Vice President, 
                     Government Programs, Cray Inc.

    Good morning, Mr. Chairman and distinguished members of the 
Committee. I am Christopher Jehn, Vice President, Government Programs 
of Cray Inc. I commend you for holding this hearing on high-performance 
computing, and I want to thank you for this opportunity to testify on 
behalf of Cray.
    Cray's rich history began in 1972, when the legendary Seymour Cray, 
the ``father of supercomputing,'' founded Cray Research. The first 
supercomputer the company built, the Cray-1, broke the world record for 
computational speed at the Los Alamos National Laboratory.
    Cray continues that tradition today. We are a global leader in 
high-performance computing, and we are the only company in the world 
solely focused on designing, building, and supporting the world's most 
powerful supercomputers.
    Our computers are purposely built to address the most demanding 
scientific and engineering problems. We give scientists and engineers 
the ability to not only get answers faster but to ask new questions at 
the frontiers of scientific discovery.
    Today, Cray's high-performance computers are addressing key 
national security missions, helping to predict severe weather, fight 
forest fires, build safer cars, discover new medicines and uncover the 
secrets to fusion power and superconductivity.
    As we discuss high-performance computing today, I want to emphasize 
four points:
    First, supercomputing is vitally important to the Federal 
Government. Federal agencies tell us this everyday. As the largest user 
of supercomputing, the Federal Government understands how necessary 
supercomputers are to fulfilling the requirements of government 
missions--from national defense and homeland security to scientific 
leadership. Agencies need supercomputing to help maintain military 
superiority, enable scientific research, advance technological 
development, and enhance industrial competitiveness. For decades, 
supercomputing has paved the way for real progress for Federal 
agencies.
    Supercomputing is also important to academic researchers and 
industry. As a key enabler for furthering science and technology, 
supercomputing has helped advance U.S. productivity and ability to 
compete in the global economy and to ultimately drive long-term 
economic growth.
    In all these areas, the need for supercomputing is growing, and to 
sustain progress as it has for decades, the Federal Government, 
academic researchers and industry must have access to increasingly more 
capable supercomputers.
    The second point I want to make is that progress in advancing 
supercomputing technology has slowed considerably. Over the last 
decade, the computer industry has standardized on commodity processors. 
With high volume low-cost processors, supercomputer clusters consisting 
of commodity parts held out a promise to users of ever-more powerful 
supercomputers at much lower cost. At the same time, the Federal 
Government dramatically reduced investments in supercomputing 
innovation, leaving the future of supercomputing in the hands of 
industry. But from industry's perspective, the supercomputing market is 
not large enough to justify significant investment in unique processor 
designs and custom interconnects--as the supercomputer market is less 
than 2 percent of the overall server marketplace, according to 
International Data Corporation. To advance supercomputing, industry has 
relied on leveraging innovation from the personal computer and server 
markets.
    Today, it has become clear that the promise of commodity-based 
supercomputers has not materialized. Because supercomputers are based 
on technology optimized for other purposes, they are exceedingly 
complex and extraordinarily difficult to use and administer. 
Computational scientists now spend enormous amounts of time, effort and 
cost modifying software algorithms to run efficiently across 
homogeneous processors. In many cases, as soon as the task is complete, 
these scientists have to repeat the process for the next-generation 
supercomputer.
    Future trends in supercomputing will only exacerbate this problem. 
Because engineers are running into physical limits trying to speed up 
individual processors in supercomputers, they are resorting to 
increasing the overall number of processors in a given system to get 
better speed. We work with hundreds to thousands of processors in 
supercomputers today. In a couple of years, we will have to work 
comfortably with tens of thousands to hundreds of thousands of 
identical processors. Since all of the processors are of the same 
architecture, further performance gains from other types of processors 
that exploit different processing models are lost. Further, as 
commodity-based supercomputers add more processors, these systems 
become less balanced as their internal commodity network becomes 
overloaded thus resulting in decreased efficiency. These systems will 
often run real-world scientific and engineering applications at only a 
small fraction of their theoretical peak capability. Most of the 
resource is wasted.
    While cheap supercomputer clusters still prove adequate for some 
applications, more and more science and engineering applications need 
better-balanced systems. That means systems with far more bandwidth and 
better reliability, cooling and power utilization, packaging, systems 
software, programming models, tools and other features than are 
available on mass-market system architectures.
    The lack of advancement in supercomputing technology not only puts 
our Nation's leadership in supercomputing at risk, but it also creates 
significant technology gaps that threaten our lead in national 
security, science and engineering, and economic competitiveness. This 
impacts the scientific and engineering community in such a way that 
many critical computational problems remain unsolvable in a timely and 
efficient manner.
    The third point I want to make is that the U.S. Government 
recognizes the importance of a healthy domestic supercomputing 
industry. A series of recent U.S. Government-commissioned studies on 
supercomputing unanimously argue for increased Federal Government 
support for supercomputer research and development. In fact, the 
Defense Department's integrated high-end computing report states, ``. . 
. many of the advantages the U.S. enjoys in technologies critical to 
national security depend to a substantial degree on the relative 
strength and diversity of its domestic commercial sources for high-end 
computing'' and recommends quadrupling Federal funding for R&D on 
supercomputing over the next 5 years.\1\ The report highlights that the 
U.S. advantage in advanced aircraft designs, ballistic missile defense 
systems, cryptanalysis, biological sciences, stealth materials, and 
many other technologies are at risk without additional Federal support 
for supercomputing R&D. The other government-sponsored reports \2\ 
delivered over the last few years also describe in more detail the 
difficulty the Federal Government faces effectively running 
applications of national importance on most of today's supercomputers. 
All of these reports call for increased Federal support for 
supercomputing.
---------------------------------------------------------------------------
    \1\ Department of Defense IHEC Report--``High-Performance Computing 
for the National Security Community.'' July 1, 2002. http://
www.hpcmo.hpc.mil/Htdocs/DOCUMENTS/041720
03_hpc_report_unclass.pdf.
    \2\ National Science Foundation report, ``Revolutionizing Science 
and Engineering Through Cyberinfrastructure: report of the National 
Science Foundation Blue-Ribbon Advisory Panel on Cyberinfrastructure.'' 
January 2003. http://www.nsf.gov/od/oci/reports/atkins.pdf.
    Classified JASON's Report examining the requirements for 
supercomputing which derive from DOE's classified weapons research. 
Fall 2003.
    Interagency High End Computing Revitalization Task Force Report--
``Federal Plan for High-End Computing.'' May 10, 2004. http://
www.nitrd.gov/pubs/2004_hecrtf/20040702_
hecrtf.pdf.
    National Research Council Report--``Getting Up to Speed: The Future 
of Supercomputing.'' November 2004. http://www7.nationalacademies.org/
cstb/project_supercomputing.html.
    The President's Information Technology Advisory Committee (PITAC) 
report--``Computational Science: Ensuring America's Competitiveness.'' 
June 2005. http://www.nitrd.gov/pitac/reports/20050609_computational/
computational.pdf.
    Joint U.S. Defense Science Board/UK Defence Scientific Advisory 
Council Task Force report--``Defense Critical Technologies.'' March 
2006.
---------------------------------------------------------------------------
    The government is doing more than just writing reports. The 
Department of Energy's Office of Science has proposed funding the 
development and deployment of a petascale computer, one capable of 
performing 1,000 trillion calculations per second. So has the National 
Science Foundation. The Department of Defense, most notably through 
DARPA's High Productivity Computing Systems (HPCS) program and the 
National Security Agency, supports research to help reinvigorate the 
advancement of supercomputing technology. For example, the goal of the 
HPCS program is to provide economically-viable next-generation 
petascale supercomputing systems for the government and industry user 
communities in the 2010 timeframe. HPCS will significantly contribute 
to DOD and industry superiority in areas such as operational weather 
and ocean forecasting, analysis of the dispersion of airborne 
contaminants, cryptanalysis, military platform analysis, stealth 
design, intelligence systems, virtual manufacturing, nanotechnology, 
and emerging biotechnology.
    My final point is that Cray is acutely aware of the current crisis 
in supercomputing. We believe we have a vision for overcoming this 
crisis. We call it adaptive supercomputing--have the machine adapt to 
the user, not the user to the machine. But we need Federal Government 
support for this vision to reach its fullest potential in a timely 
manner, as the market is not large enough to fund the risky, leading-
edge research and development that is required.
    Cray's vision of adaptive supercomputing grew out of its 
partnership with the DARPA HPCS program. Cray, in collaboration with 
AMD, has proposed a paradigm shift in the supercomputing industry that 
will enable the building of much more powerful, yet significantly 
easier to use supercomputers than are built today. Like all previous 
Cray computers, the new supercomputers will be designed from the bottom 
up rather than be based on a collection of PC and server commodity 
parts.
    Using revolutionary technology, future Cray supercomputers will 
employ diverse microprocessor architectures that can dynamically adapt 
to scientific requirements in a transparent, scalable, robust and 
optimized way. This will allow computational scientists to focus on 
their unique scientific problems and application requirements instead 
of being forced to conform to the supercomputer. Systems will be 
radically easier to program, much more broadly applicable, and more 
resistant to failure. They will give scientists and engineers the tool 
they need to solve the multi-scale, multi-physics problems of the 
future. Computational scientists will experience a tremendous 
productivity boost saving government and industry time and money while 
enhancing competitiveness.

Recommendations
    Our recommendation to this Committee and the Congress is to fully 
fund the Administration's proposed government investments in 
supercomputing. This includes funding supercomputing programs in the 
Department of Energy, the National Science Foundation, the National 
Aeronautics and Space Administration, and within the Department of 
Defense. To continue international leadership in science, industry and 
national security, the U.S. Government must fully fund the continued 
evolution of supercomputers and give scientists access to the 
computational capability for a wide range of scientific and engineering 
disciplines. This investment will be justified by an array of future 
breakthroughs from more efficient, quieter planes and space vehicles to 
improvements in digital imaging and drug discovery. The promises of 
supercomputers are limited only by our imagination.
    For its part, the Administration should build on its recent 
initiatives and develop and fund a coordinated research and development 
program for supercomputing as recommended in the many reports cited 
above. The Administration should also take a stronger leadership role 
in persuading other Federal agencies to make use of supercomputing and 
computational science to carry out agency missions. Many agencies have 
realized only limited scientific progress because they are reluctant to 
complement experiment-based science with computational science. The 
Administration should identify gaps in computational science usage and 
develop programs to close these gaps.
    We also want to express our support for H.R. 28, the High 
Performance Computing Revitalization Act of 2005. We worked with the 
House Science Committee on this bill. It not only updates current law, 
but it reemphasizes the need for continued advances in supercomputing.
    In conclusion, I would like to laud both the Administration's and 
Congress's leadership with respect to high-performance computing. 
Recent developments have been very encouraging. Both Congress and the 
Administration are seeing high-performance computing as a key enabler, 
even a catalyst for pushing out the frontiers of science and 
technology. In the report ``Rising Above the Gathering Storm'' \3\ the 
National Academies of Science stated: ``The committee is deeply 
concerned that the scientific and technical building blocks of our 
economic leadership are eroding at a time when many other nations are 
gathering strength.'' Supercomputing is one of those key building 
blocks. The Japanese and Chinese governments recognize this and have 
taken significant steps to boost supercomputing activities 
domestically. They see what we see. What supercomputers have done for 
us today will pale in comparison to what supercomputers will do for us 
tomorrow. Now is the time to invest.
---------------------------------------------------------------------------
    \3\ National Academies of Science, ``Rising Above the Gathering 
Storm: Energizing and Employing America for a Brighter Economic 
Future,'' 2006, http://newton.nap.edu/catalog/11463.html. 

   STATEMENT OF JACK WATERS, EXECUTIVE VICE PRESIDENT/CHIEF 
          TECHNOLOGY OFFICER, LEVEL (3) COMMUNICATIONS

    Mr. Waters. Mr. Chairman, Senator Cantwell, my name is Jack 
Waters and I'm Executive Vice President and Chief Technology 
Officer of Level (3) Communications. Thank you for the 
opportunity to testify today on behalf of Level 3. We believe 
that high-performance networking is an essential element of 
high-performance computing, critical to our Nation's 
competitiveness, and should be a central part of Federal policy 
regarding the Nation's cyber-infrastructure.
    With that, I am going to turn to a presentation that 
summarizes my testimony.



    I would like to make two points. The first is that high-
performance networking is a key component of high-performance 
computing and the second is that the government has played, and 
should continue to play, a leadership role in both networking 
and computing.
    Next slide.

    
    
    I will tell you very briefly about Level 3. Level 3 is a 
little bit new, a new company to the telecom world. We were 
founded about 9 years ago and we constructed a network in the 
U.S., across the U.S., and in Europe. We had the idea in mind 
at the time that, what seems obvious today, Internet technology 
and optical technologies would be the waves of the future and 
we should optimize our company around those technologies.
    Next slide.

    
    
    The point of this slide is really to depict that for years 
we have considered computing information and storing 
information as the only variables in how we handle information 
as an industry, and recently the economics of moving 
information around on a network have become compelling, and 
that is why networking should be such a key component of high-
performance computing.
    If you turn to the next slide.

    
    
    The Federal Government has played a pretty key role and has 
made some important investments over the last couple of 
decades, and I would like to point out three: the original 
investment in the ARPAnet and its communication protocol, 
TCPIP, the language that computers speak to each other over 
networks; the original funding in the NSF supercomputer centers 
and the network that interconnects those supercomputer centers.
    Well, what did we get out of those investments? It is 
pretty daunting when you think those original investments led 
to the commercial Internet as we know today in both our 
personal and professional lives, and, even more basic, the 
browser. The browser actually came from one of the NSF-funded 
supercomputer centers, NCSA. It was called Mosaic and it is 
probably something that everyone in this room and around the 
world uses every day as part of their Internet use.
    If you turn to the next slide.

    
    
    Those investments have led to the growth that I know that 
we all have seen. This picture actually shows how incredible it 
has been and frankly continues to be.
    If you turn to the next slide.

    
    
    I would like to point out a couple of particular research 
projects. CERN is a particle physics laboratory in Switzerland 
and they are building a machine called the Large Hadron 
Collider. It costs about $8 billion. After it is finished, it 
is going to produce about 1,500 times the information held in 
the Library of Congress, so a fair amount of information, and 
2,000 scientists around the world in 31 countries, including 
our country, need access to that information. The only way to 
provide access to that information is through high-performance 
networking.
    If you turn to the next slide.

    
    
    There is another initiative closer to home, called 
``Internet2,'' a not-for-profit company that is focused on 
providing networking to the R&E community. They have recently 
announced plans to design and deploy a new nationwide 
infrastructure. Once completed, this infrastructure will allow 
researchers an unparalleled access to computing and storage 
facilities across our country.
    If you turn to the next slide.

    
    
    You will see what we get from this investment and 
Internet2's broad reach which touches many institutions across 
the country. At the end, users will be able to dynamically 
tailor their networking resources and needs to suit their 
research initiatives.
    If you turn to the next slide----

    
    
    --I thought I would take a moment and give you an idea of 
what these investments have given us as far as a figure of 
merit. The figure of merit we networking people love to use is 
the amount of information in the Library of Congress and how 
long it would take to transfer it. If you looked 20 years ago, 
at the original NSFNet, it would have taken about 50 years on 
that network to transfer that information. If you looked at the 
investment that Internet2 is making this year in the 
infrastructure that is being built, it would take about 15 
minutes to transfer that information. So the investments have 
garnered great capability in networking.



    In summary, we believe that Federal policy should achieve a 
balance and focus on three key elements: computing, software, 
and what I have spoken about today, networking; a balanced 
approach will lead to American innovation, facilitate advanced 
research, contribute to our homeland security and national 
defense, and fortify our competitive position.



    Thank you very much.
    [The prepared statement of Mr. Waters follows:]

   Prepared Statement of Jack Waters, Executive Vice President/Chief 
              Technology Officer, Level (3) Communications

    Mr. Chairman and Members of the Subcommittee, my name is Jack 
Waters, and I am Executive Vice President and Chief Technology Officer 
of Level (3) Communications. Thank you for the opportunity to testify 
today on behalf of Level (3) Communications. We believe that high-
performance networking is an essential element of high-performance 
computing, critical to our Nation's competitiveness and should be a 
central part of Federal policy regarding the Nation's cyber-
infrastructure.
    First, let me commend the Subcommittee for its work in approving 
the American Innovation and Competitiveness Act (S. 2802). I share the 
Subcommittee's view that a renewed commitment to basic research will go 
a long way to ensuring the competitiveness of the United States and to 
maintaining and improving the United States' innovation in the 21st 
century by: increasing research investment, increasing science and 
technology talent, and developing the Nation's ``innovation 
infrastructure.'' I believe that high-performance computing and the 
high-performance networks that interconnect and facilitate information-
sharing between the high-performance computing centers are key elements 
of the Nation's ``innovation infrastructure,'' and are essential to 
ensuring our homeland security, the strength of our national defense, 
and ultimately, our continued economic competitiveness in the global 
economy.
    Level (3) Communications is a U.S. company focusing on 
international communications and infrastructure services. Our network 
has more than 36,000 fiber route miles and provides high-bandwidth 
services in 15 countries. Level (3) Communications constructed and now 
operates one of the largest Internet Protocol (IP) and optical 
transport backbone networks in the United States and Europe, utilizing 
the latest fiber and optical technologies. Level 3 is regarded as one 
of the most technologically advanced carriers in the world, recognized 
by the Smithsonian Institution for building ``The world's first 
upgradeable international fiber optic network to be completely 
optimized for Internet Protocol technology . . .''
    The Federal Government has played a vital role in both high-
performance computing and high performance networking for several 
decades. In 1979, after the successful deployment of the ARPAnet 
(originally a military network funded by the Department of Defense) the 
National Science Foundation saw the need to link computer science 
researchers across the Nation and funded a basic network called CSnet. 
In 1984, with several advancements in high-performance computing 
occurring, the NSF funded the construction of five supercomputer 
centers across the country and connected these centers with a network 
called the NSFnet. Although more than twenty years ago, these 
investments, along with subsequent others by the Federal Government, 
have helped drive many technology innovations that all of us use every 
day. A few examples will help me illustrate my point.
    Today, we all know and use the Internet in both our personal and 
professional lives. The NSFnet mentioned previously, was a key piece of 
the early infrastructure that started it all. This network which 
interconnected 5 supercomputer centers in 1985 and 50,000 networks in 
1995, the time of its decommissioning, was the platform on which the 
commercial Internet that we know today was founded.
    Another key piece of Internet technology came from one of the five 
supercomputer centers that the NSF funded. Although Tim Berners-Lee is 
quite rightly credited with the idea of the World Wide Web, the first 
widely used Internet browser was developed at the University of 
Illinois' National Center for Supercomputing Applications. This 
browser, named Mosaic, became an overnight success allowing early 
Internet users the ability to find information across the vast global 
network. All of this happened in 1993, many years before the world 
really understood what the Internet would be, through our government's 
foresight and financial support.
    Increasingly, advanced research in the United States and around the 
globe is accomplished collaboratively by researchers and data sources 
which are geographically distributed. The quantities of empirical and 
higher-order data used in this research are also increasing at an 
incredible pace. As such, the need to share large quantities of 
information in a timely manner among geographically distributed 
research centers becomes an essential part of accomplishing the 
objectives of these advanced research programs. Let me use several 
examples to illustrate this point:
    The Large Hadron Collider (LHC), located at the European particle 
physics research center, CERN (Conseil Europeen pour la Recherche 
Nucleaire), cost approximately $8 billion to construct and is planned 
to begin operation in 2007. Once online, the Collider will produce an 
output stream of data approaching a Terabit (one trillion bits) per 
second which will be shared with 34 research centers around the world. 
The existing network infrastructure is not sufficient to handle this 
demand.
    In the field of medical research, the newest 1.25 MeV (Mega-
electron volts) ultra high-voltage electron microscopes, which allow 
detailed structural studies of biological specimens, produce network 
bandwidth requirements that approach 100 Gigabits per second--a 
requirement equivalent to the capacity planned for the largest American 
research network, Internet2 now under construction.
    Today the Tera-grid network, which recently received increased 
funding from the National Science Foundation, links seven U.S.-based 
supercomputing and research centers. Tera-grid has 200 teraflops (one 
trillion floating point operations per second) of computational 
capacity, 20 terabytes of storage and will reach sustained data flows 
between these centers approaching and eventually exceeding 1 terabit 
per second.
    As the first two examples illustrate, the basic instrumentation in 
advanced research can be so costly that simple economics mandate that 
these essential elements be shared by the many research centers and 
scientists rather than duplicating the basic functions. Further, all of 
the examples demonstrate the trend toward distributed use of enormous 
quantities of basic research data. Increasingly, refined specialty and 
inter-disciplinary research initiatives also create an increasing need 
for collaboration among various research centers and inter-disciplinary 
research teams. These two factors, cost-efficiency and the need for 
research collaboration among geographically distributed centers, 
underlie and motivate the need for efficient, high-performance networks 
to interconnect these various research centers.
    A final case in point is the National Science Foundation's (NSF) 
Major Research Engineering Facility Construction (MREFC) Program, which 
provides funding for complex research instruments at 10 centers across 
the United States, plus one in Antarctica. Each of these centers has an 
instrumentation and discipline-specific focus--such as ecology, 
physics, magnetism, etc. The basic data produced by these instruments 
are shared among the scientific community by manual transference of 
data or, more efficiently, across networks which can speed the 
researchers' access to these basic data streams.
    It is clear that the Federal Government has historically recognized 
the need to fund both high-performance computing and high performance 
networking. The investments made two decades ago have left a proud 
legacy for the benefit of the entire world. It is also clear that this 
Subcommittee and the Federal Government recognize the need for 
continued funding and research in the network component of the Nation's 
``cyber-infrastructure'' and have taken important steps to address 
these issues. Examples include:

        In 2003, the NSF Blue Ribbon Advisory Panel, published a report 
        entitled ``Revolutionizing Science and Engineering through 
        cyber-infrastructure'' in which it stated, ``High-speed 
        networks are a critical cyber-infrastructure facilitating 
        access to the large, geographically distributed computing 
        resources, data repositories, and digital libraries. As the 
        commodity Internet is clearly not up to the task for high-end 
        science and engineering applications, especially where there is 
        a real-time element (e.g., remote instrumentation and 
        collaboration), a high-speed research network should be 
        established and the current connections program extended to 
        support access to this backbone as well as to provide 
        international connections.''

        The National Science and Technology Council in 2004 called for 
        achieving aggressive networking goals such as:

           networks with 1,000 times existing capabilities,

           with better security and trust mechanisms, and

            development of inter-optical networks and 
        middleware to couple networks with software.

        The National Science Foundation's recently announced plans for 
        the Global Environment for Network Innovations (GENI) with the 
        primary goal to enable the research community to invent and 
        demonstrate a global communications network and related 
        services that will be qualitatively better than today's 
        Internet.

    In addition to these important Federal initiatives is the work of 
our Nation's research universities. Recently, the nonprofit consortium 
known as Internet2, serving more than 200 research universities, took 
an important step toward meeting the growing bandwidth requirements of 
many of the United States' top research centers.
    On June 15, 2006, Level 3 and Internet2 announced an agreement to 
design and deploy a new nationwide network and new services to enhance 
and support the advanced needs of the academic and research community. 
Internet2's new network will provide its members with 100 gigabits per 
second (Gbps) of network capacity between key research centers, more 
than 10 times that of its current backbone. Even with this big step 
forward, Internent2's members have asked Level 3 to provide a network 
platform capable of handling even larger bandwidth demands. 
Accordingly, a key design characteristic of this network is the ability 
to quickly scale to add capacity as members' requirements evolve.

        [Visual Aid]: This map represents a small fraction of the 
        Institutions who are members of Internet2. This illustration 
        also shows a number of the federally-funded research and 
        development centers which will directly or indirectly benefit 
        from the Internet2 backbone.

        
        
    Under the agreement with Internet2, Level 3 will deploy leading 
edge digital optical technologies to provide multiple ten (10) Gbps 
wavelengths and enable rapid bandwidth provisioning across the entire 
network. These new optical services will allow researchers and 
scientists to obtain dedicated one (1) Gbps sub-wavelengths or entire 
ten (10) Gbps wavelengths and optimize the utilization of the network 
to suit the information-sharing needs of the researchers.
    In addition to providing high capacity, scalable bandwidth, 
achieving efficient utilization of the network is critical to ensuring 
that researchers have the bandwidth they need when they need it. 
Optimal utilization of network resources improves the economic 
efficiency of the research, allowing more robust and dynamic use of the 
network. Internet2's network and the flexible optical services it 
provides, will increase flexibility and support bandwidth-intensive 
experimental applications which have direct impact on the United 
States' research agenda, homeland security, national defense and our 
economy. Like the Federal initiatives cited earlier, Internet2 
demonstrates that the network is an essential component of the Nation's 
``cyber-infrastructure'' and essential to achieving the objectives of 
the most advanced research being conducted in the United States and 
abroad.

Summary and Recommendations
    In summary, I believe that a Federal policy that achieves a balance 
of investment and focus on the three key elements of the Nation's 
``cyber-infrastructure''--computing power, software, and networking--is 
likely to yield the greatest benefits. A balanced approach will: (1) 
contribute to the attainment of the goals of the American Innovation 
Act; (2) work to ensure that all of the essential elements of the 
Nation's ``innovation infrastructure'' are available to facilitate 
advanced research; (3) contribute to homeland security and national 
defense; and (4) fortify the United States' economic and technological 
competitive position.
    I thank you for your time and I am happy to answer any questions 
you have.

    Senator Ensign. Thank you.
    Mr. Lombardo.

           STATEMENT OF JOSEPH M. LOMBARDO, DIRECTOR,

         NATIONAL SUPERCOMPUTING CENTER FOR ENERGY AND

        THE ENVIRONMENT, UNIVERSITY OF NEVADA, LAS VEGAS

    Mr. Lombardo. Thank you for the opportunity to appear 
before the Committee and offer observations on the role of 
Federal policy in the area of high-performance computing as it 
relates to academic research. My name is Joseph Lombardo, 
Director of the National Supercomputing Center for Energy and 
the Environment. The Center is a mature high-performance 
computing center located at the University of Nevada, Las 
Vegas. The Center was established in 1989 and has played an 
important role in the high-performance computing community by 
providing a resource for academic researchers in the fields of 
energy and the environment and has an impressive track record 
of sponsored research from a range of Federal agencies, 
including the Department of Energy, the Department of Defense, 
Interior, EPA, Health and Human Services, NOAA, NSF, and 
others.
    I would like to make the following observations. The 
history of Federal support for high-performance computing has 
been tied to perceptions that high-end computing is crucial to 
a broad definition of national security. That is, that the 
strength of the U.S. is tied not only to military hardware, but 
to scientific and technological preeminence. High-performance 
computing is crucial to that preeminence, as it is a basic tool 
for the advanced research across many fields.
    Initial Federal support came with the Lax report issued in 
1983 by the National Science Foundation, which perceived that 
the Japanese sixth generation computer would give Japan a large 
lead over the U.S. in high-end computing. The Lax report 
recommended Federal funding for supercomputing centers in open 
environments such as universities and for training, software 
engineering, and related activities.
    This report led to the formation of the NSF centers as well 
as other Federal and state-funded centers across the country. 
In the late 1980s and early 1990s, high-performance computing 
funding was directed through collaborative relationships 
between government, corporate sector, and academic research 
consortia, leading to the formation of policies that 
established a national priority list of Grand Challenges in 
research, addressed through high-performance computing tools. 
This era saw broadening of the high-performance computing 
manufacturing base, as well as significant software and tool 
development. This might be considered the highwater mark of 
Federal interest and funding for high-performance computing.
    Beginning in 1993, Federal policy reversed and deemphasized 
Grand Challenge problems. Grand Challenge problems are 
extremely difficult to solve, requiring several orders of 
magnitude improvement in computational capability. The focus 
then shifted to distributed computing and moved toward 
commercial off-the-shelf technology. Such initiatives led to a 
broader range of individuals working in the scientific 
computing, but basically starved the high end of the high-
performance computing field.
    At the end of the 1990s, DARPA and other organizations 
began to see that foreign countries such as the Asian groups 
were overtaking the U.S. position in high-performance computing 
once again and recommended policies that would fund and support 
the high end of the field again. The DARPA High Productivity 
Computing Systems program is a good example of this shift back 
toward an emphasis on high-end capability. The DARPA program is 
focused on providing a new generation of economically viable 
high-productivity computing systems for the national security 
and industrial user community in the 2010 timeframe.
    This trend has continued with the High Performance 
Computing Revitalization Act, the President's 2006 State of the 
Union Address, and with the Fiscal Year 2007 budget, which 
increased DOE's high-performance computing programs by almost 
$100 million.
    This brief recounting of the history of Federal support for 
high-performance computing demonstrates that national interest, 
academia, and high-performance computing communities are joined 
at the hip. Scientific and technological preeminence for the 
U.S. is related directly to high-performance computing. Support 
for Federal funding of high-performance computing has ebbed and 
flowed as a result of perceived foreign competition. 
Collaborations of Federal laboratories and agencies, academic 
institutions, and corporate interests are key to advancing both 
technologies and applications, but require Federal funding to 
do so.
    Based on the above, I would like to make the following 
observations: One, Federal policy should recognize high-
performance computing as vital to the scientific and 
technological strength of the U.S. and as such should be 
considered as crucial to national security.
    Two, Federal funding for high-performance computing should 
encourage development of cutting edge high-end technologies 
capable of addressing the Grand Challenge problems as well as 
the mid-range problems.
    Three, Federal policy should encourage expansion of 
applications in fields where high-performance computing is not 
yet a core research tool, as an example agriculture, many of 
the biomedical areas, and transportation.
    Four, Federal policies and funding should be allocated to 
encourage a diverse industry with a range of companies given 
opportunity to develop and deploy their technologies. Such 
broad applications and procurements are crucial to sustain a 
viable high-performance computing manufacturing community not 
dominated by a single corporate interest.
    Thank you for your interest and for the opportunity to 
share my thoughts with the panel.
    [The prepared statement of Mr. Lombardo follows:]

     Prepared Statement of Joseph M. Lombardo, Director, National 
  Supercomputing Center for Energy and the Environment, University of 
                           Nevada, Las Vegas

Introduction
    Thank you for the opportunity to appear before the Committee and 
offer observations on the role of Federal policy in the area of High-
Performance Computing as it relates to academic research.
    My name is Joseph Lombardo, Director of the National Supercomputing 
Center for Energy and the Environment--the center is a mature High-
Performance-Computing Center located at the University of Nevada Las 
Vegas. The center was established in 1989 and has played an important 
role in the High-Performance Computing community by providing a 
resource for academic researchers in the fields of Energy and the 
Environment, and has an impressive track record of sponsored research 
from a range of Federal agencies, including the Department of Energy, 
Department of Defense, Interior, EPA, Health & Human Services, NOAA, 
NSF and others.
    I'd like to make the following observations:
    The history of Federal support for High-Performance Computing has 
been tied to perceptions that high-end computing is crucial to a broad 
definition of national security--that is, that the strength of the U.S. 
is tied not only to military hardware but to scientific and 
technological preeminence. High-performance computing is crucial to 
that preeminence, as it is a basic tool for advanced research across 
many fields.
    Initial Federal support came with the Lax Report, issued in 1983 by 
the National Science Foundation, which perceived that the Japanese 6th 
generation computer would give Japan a large lead over the U.S. in 
high-end computing. The Lax Report recommended Federal funding for 
supercomputing centers in open environments, such as universities, and 
for training, software engineering, and related activities. This report 
led to the formation of the NSF centers as well as other Federal and 
state-funded centers across the country.
    In the late 1980s and early 1990s, High-Performance Computing 
funding was directed through collaborative relationships between 
government, corporate sector, and academic research consortia, leading 
to the formation of policies that established a national priority list 
of ``Grand Challenges'' in research, addressed through High-Performance 
Computing tools. This era saw broadening of the High-Performance 
Computing manufacturing base, as well as significant software and tool 
development. This might be considered the ``highwater mark'' of Federal 
interest and funding for High-Performance Computing.
    Beginning in 1993, Federal policy reversed and deemphasized ``Grand 
Challenge'' problems. Grand Challenge problems are extremely difficult 
to solve, requiring several orders-of-magnitude improvement in 
computational capability. The focus shifted to distribute computing and 
moved toward ``commercial off-the-shelf'' technology. Such initiatives 
led to a broader range of individuals working in scientific computing, 
but basically starved the high end of the High-Performance Computing 
field.
    At the end of the 1990s DARPA and other organizations began to see 
that foreign countries, such as Asian groups, were overtaking the U.S. 
position in High-Performance Computing once again, and recommended 
policies that would fund and support the high end of the field once 
again. The DARPA High Productivity Computing Systems program is a good 
example of this shift back toward an emphasis on high-end capability. 
The DARPA program is focused on providing a new generation of 
economically-viable high productivity computing systems for the 
national security and industrial user community in the 2010 timeframe. 
This trend has continued with the High-Performance-Computing 
Revitalization Act, the President's 2006 State of the Union Address, 
and with the FY07 Budget which increased DOE's High-Performance 
Computing programs by almost $100 million.
    This brief recounting of the history of Federal support for High-
Performance-Computing demonstrates that national interest, academia, 
and the High-Performance-Computing community are joined at the hip.
    Scientific and technological preeminence for the U.S. is related 
directly to High-Performance-Computing. Support for Federal funding of 
High-Performance-Computing has ebbed and flowed as a result of 
perceived foreign competition. Collaborations of Federal laboratories 
and agencies, academic institutions and corporate interests are key to 
advancing both technologies and applications, but require Federal 
funding to do so.
    Based on the above, I would make note of the following 
observations:

        1. Federal policies should recognize High-Performance Computing 
        as vital to the scientific and technological strength of the 
        U.S. and as such, should be considered as crucial to national 
        security.

        2. Federal funding for High-Performance Computing should 
        encourage development of cutting edge, high-end technologies, 
        capable of addressing ``Grand Challenge'' problems as well as 
        mid-range projects.

        3. Federal policies should encourage expansion of applications 
        in fields where High-Performance Computing is not yet a core 
        research tool--e.g., agriculture, many bio-medical areas, and 
        transportation.

        4. Federal policies and funding should be allocated to 
        encourage a diverse industry, with a range of companies given 
        opportunity to develop and deploy their technologies. Such 
        broad applications and procurements are crucial to sustain a 
        viable High-Performance Computing manufacturing community not 
        dominated by a single, corporate interest.

    Thank you for your interest and for the opportunity to share my 
thoughts with the panel. I would be pleased to answer questions the 
members may have.

    Senator Ensign. Thank you.
    Mr. Garrett.

 STATEMENT OF MICHAEL GARRETT, DIRECTOR, AIRPLANE PERFORMANCE, 
                  BOEING COMMERCIAL AIRPLANES

    Mr. Garrett. Mr. Chairman, Senator Cantwell, Good morning. 
My name is Michael Garrett, Director of Airplane Performance 
for the Boeing Commercial Airplane Company in Seattle, 
Washington. In that role I am responsible for the performance 
characteristics of all our commercial products, including the 
new product development such as on the 787. That includes the 
mission performance capabilities such as fuel burn and range, 
as well as the environmental performance and capability of our 
products, such as noise characteristics and emissions.
    Today I am going to give you a brief summary of the role of 
supercomputing, of high-performance computing at the Boeing 
Company. Let me get this up real quick.
    Senator Cantwell. It is hard when you have to testify and 
run your own demo.
    Mr. Garrett. Thank you.
    The first one I want to talk about is scope. High-
performance computing is not only used on our commercial 
transport aircraft, but on our military aircraft, launch 
vehicles, and our space vehicles as well. It plays a 
significant role and has been in the development and design of 
these products.
    With respect to the impact of performance computing, high-
performance computing, when it is connected with our 
computational tools and methods, it kind of supercharges the 
design process. They provide solutions much faster than we have 
ever been able to do before, to much more complex problems, 
more accurate solutions in that we are able to predict better 
and lower the risk of the development of our products because 
we can predict how the airplanes are going to perform when they 
deliver to our customers better than we ever had before.
    We also are able to enhance the safety and the 
environmental aspects of our airplanes, so that they deliver to 
our customers what they are looking for.
    On the business side, we have significantly reduced the 
cycle time as we develop our products and lower the costs. 
Basically, high-performance computing allows us to validate 
technology and get it to the market faster, which results in 
lower costs and better performance of our products for our 
customers.
    So I want to give one current example of high-performance 
computing and what it has done. This is in the area of noise 
reduction. The picture on the lower right shows an application 
of technology on a 777 flight test airplane. It shows what are 
called chevrons, these saw-tooth like structures, which is 
typically a straight leading edge of the nacelles. We have 
looked at a technology that allows us to reduce noise on the 
787 and by doing so and making it the lowest noise airplane in 
its class, from an environmental standpoint lowering the noise 
in the areas we live and work.
    We did not get there without high-performance computing and 
our analysis tools, which are shown in the upper left and in 
the lower left. Those are simulations that allow us to run a 
number of different configurations very quickly before ever 
going into the acoustic tunnel to test and before we ever go to 
flight test. So it reduces the cost because the cost of the 
wind tunnel goes up and the cost of flight test obviously is 
going up with the price of fuel. The more we can do to 
simulate, the better we are.
    The future. What we show here is the ability that we need 
in the future to look at the acoustics of the entire airplane, 
not just of the engine. The model you see there that is 
represented from our computational fluid codes uses high-
performance computing to do a very complex type of problem with 
separated flows, looking at the gear down, at the flaps down. 
It is time-based because it is separated flow, so it is time-
dependent as well as running the engines at the same time.
    This solution takes hours. We need to get this down to 
seconds because this is only one solution. We need multiple 
solutions to handle all the flight conditions that are needed, 
and this is where the future is.
    So in summary, four major points. High-performance 
computing is used throughout the Boeing Company from a product 
development standpoint and we will be using it more and more in 
the future as well. It allows us to meet our business case 
conditions by reducing our cycle time and getting our products 
to the market faster. We are investing in high-performance 
computing now primarily because of the 787. On an annualized 
basis, we have increased our investment 50 percent year over 
year the last 7 years. We have invested tens of millions of 
dollars in the development of our new products.
    Last, the continued improvements in high-performance 
computing through faster and more efficient computers will 
enable Boeing to provide more efficient and more capable 
products to our customers at reduced cost.
    Thank you.
    [The prepared statement of Mr. Garrett follows:]

Prepared Statement of Michael Garrett, Director, Airplane Performance, 
                      Boeing Commercial Airplanes

Introduction
    Good morning, Mr. Chairman and Members of the Subcommittee. I am 
Michael Garrett, Director, Airplane Performance for the Commercial 
Airplane Division of the Boeing Company. I have worked at Boeing and 
McDonnell Douglas for 27 years with a broad range of experiences in 
product development, program management and marketing. In my current 
position, I have responsibility for the overall performance 
characteristics of all Boeing Commercial Airplanes, including new 
products, such as the 787.
    At Boeing, we pride ourselves for understanding our customers' 
requirements and then developing, designing, building and delivering 
airplanes that meet or exceed those requirements.



    High-performance computing has fundamentally changed the way that 
Boeing designs flight vehicles, whether it be commercial transports, 
military fighters, unpiloted aircraft, guided bombs, launch vehicles, 
or crewed-space exploration vehicles. Computational tools are being 
used to create numerical simulations to assess system performance--
replacing the more costly and time consuming requirements for physical 
testing. For example, these new tools are being used to determine the 
aerodynamic performance of entire airplanes, the optimum structural 
layout to minimize weight, and the radar cross-section of a stealthy 
vehicle. It is the evolution of computing hardware that has enabled 
more efficient simulations with reductions in overall design cycle 
times.

High-Performance Computing Is Good Business



    When we combine our computational design tools with the high-
performance computing resources, we obtain incredible efficiencies in 
the design processes we use to develop our airplanes. Complex processes 
and simulations, such as computational fluid dynamics, can be run much 
more quickly, at lower cost, and at a level of fidelity and accuracy 
that is equal to that achieved in physical vehicle testing. While we 
will never eliminate wind tunnel and flight testing, more powerful 
computing tools allow us to better predict results, therefore reducing 
technical risk, while reducing physical testing costs.

High-Performance Computing in Computation Fluid Dynamics 
        Applications

        
        
    One of the best utilizations of high-performance computing is in 
the development of computational fluid dynamics (CFD). While CFD has 
been in use at Boeing for 30 years, the most extensive application has 
been on our newest commercial aircraft, the 787 Dreamliner. The use of 
CFD tools has allowed Boeing engineers to address a wide range of 
design challenges, including traditional wing design, the even 
distribution of cabin air and reduction in overall airplane noise.

High-Performance Computing in Aircraft Noise Reduction



    High-performance computing, together with our CFD tools, has also 
played a significant role in reducing airplane noise. The example above 
shows how an engine noise-reduction feature called ``chevrons'' was 
developed for application to our commercial airplanes. The 787 will be 
the first Boeing airplane with this technology. We were able to 
simulate the noise reduction characteristics of numerous chevron 
configurations and determine the best configuration for noise reduction 
before ever testing in the acoustic tunnel or in actual flight test. 
This means the 787 will be a quieter aircraft, making it more 
environmentally-friendly for those who live and work near airports.
    It is the knowledge gained from the this process that reduces the 
product development life-cycle allowing our customers to get our 
products faster while meeting our commitment to improve the 
environmental performance of our airplanes.

High-Performance Computing in Wing Design



    Another example of the benefit of high-performance computing and 
improved computational capability has been in our wing development over 
the last 25 years. In 1980, Boeing tested 77 wings in wind tunnels to 
arrive at the final configuration of the 767. Just 25 years later, we 
built and tested 11 wings for the 787--a reduction of over 80 percent. 
Those 11 wings took less people to design, less time to design, and the 
wind tunnel results matched the CFD predictions.
    While our CFD tools today are very good, there are still some 
flight conditions that cannot be simulated very well. These conditions 
will continue to require significant wind tunnel testing. As more 
advanced computer hardware is developed, the computational tools and 
processes should improve and we will one day be able to calculate the 
airplane's characteristics everywhere in the flight regime with high 
fidelity.



    The chart above shows another wing design application which 
resulted in a configuration that could not have been designed without 
CFD design tools and high-performance computing. The Air Force has a 
requirement for a battlefield delivery transport. It must operate out 
of unimproved, very short runways--runways shorter than C-17s can use 
today. Meeting these challenging specifications requires a new and 
innovative wing with performance never previously demonstrated. A new 
active flow control technology was evaluated to achieve that 
performance using CFD. This computation, shown in the pictures above, 
demonstrated that when air remains attached to the wing, as in the 
picture in the lower right hand corner, lift is increased. This 
additional lift enables the aircraft to take-off and land in shorter 
distances. After computer simulation, this concept was then 
successfully demonstrated on the Advanced Tactical Transport model at 
the NASA Langley wind tunnel.

Future of High-Performance Computing



    As previously stated, we have reduced the amount of wind tunnel 
testing required for new product development. Our vision for a future 
design environment would be that all simulation work would be done 
computationally, enabled through more powerful high-performance 
computing tools. This would allow us to test only two or three wings in 
the wind tunnel versus the 11 for the 787. This will not only 
dramatically reduce the non-recurring cost to develop an airplane but 
also reduce the time it takes to bring a new product to market. Instead 
of developing a new airplane once a decade, we can potentially develop 
one in significantly less time, allowing us to be more responsive to 
market demand.



    An even greater challenge lies in the area of acoustics. Today 
laboratory and/or flight tests must be conducted to determine the 
acceptability of candidate airplane configurations to meet community 
noise requirements. In the future we hope to do all the simulations 
within the computer.
    This is a problem that is probably decades away from being 
addressable because noise covers a wide frequency. The numeric grid to 
capture the shorter wavelengths drives up the size of the problem 
dramatically--as does the requirement to model the landing gear, all 
flap and slat details, the engine (running!)--and it is all time-
dependent. As the hardware continues to improve, we will incrementally 
work our way up to meeting this challenge.

High-Performance Computing Usage at Boeing



    Boeing is committing large amounts of resources to provide the 
necessary computing capability we require. During the development of 
the 787, we have nearly doubled the capacity of our high-performance 
computing data center year after year. This is a big investment of 
capital, but one that we are willing to make because there is a 
measurable return for that investment. While our high-performance 
computing usage has increased, the cost per unit has been dramatically 
reduced by 50 percent making our development tools more and more cost 
effective.

Conclusions



    Boeing has made extensive use of high-performance computing in 
addressing a wide range of issues across all of its products. While 
high-performance computing is a valuable tool across the entire product 
cycle, its primary contribution has been in technology validation and 
its application into new product development.
    Our reliance on High-Performance Computing continues to grow as 
better, faster and more cost effective processing is available. This 
will enable Boeing to deliver better value to our customers through 
products that are more efficient and capable at significantly lower 
cost.
    Again, Mr. Chairman, I appreciate this opportunity to testify 
before the Subcommittee.

    Senator Ensign. Thank you.
    Dr. Burt.

    STATEMENT OF STANLEY K. BURT, Ph.D., DIRECTOR, ADVANCED 
                  BIOMEDICAL COMPUTING CENTER

    Dr. Burt. Thank you, Mr. Chairman, for allowing me to 
testify today. My written report has many examples of high-
performance computing applied to biology and computational 
bottlenecks. I will limit myself to just a few remarks relating 
to biology, high-performance computing, and biomedical 
research.
    I am the Director of the Advanced Biomedical Computing 
Center, ABCC, which is the principal high-performance, high-
capacity computing resource for the National Cancer Institute 
of the National Institutes of Health. The ABCC was founded in 
1986, when its first supercomputer came in. We provide 
researchers high-performance computing tools for their research 
in the complex field of cancer. The goal of the ABCC is to 
provide the cyberstructure for data-intensive computing. And 
unlike some other supercomputing centers, the ABCC only 
supports biological research.
    Both molecular biology, which is the basic science, and 
oncology research in high-performance computing have advanced 
dramatically over the past decade. In fact, high-performance 
computing has emerged as a basic tool to address very complex 
issues, commonly referred to as ``Grand Challenges,'' in most 
areas of research. Computer simulations in fields such as 
physics and chemistry have become the third leg of the stool 
along with theory and experiment.
    As molecular biology has advanced rapidly, there has not 
been the adoption of HPC as a necessary tool and the simulation 
of biological problems has not in general been carried out by 
true molecular biologists. Applications of HPC in medicine and 
the biosciences have lagged behind the physical sciences. While 
Boeing designs its new airplanes using HPC technology, all too 
many molecular biologists are still at the bench using familiar 
approaches to address the very complex issues and identifying 
and assessing normal and abnormal cell structures that are the 
basis of cancer research.
    There is the need for biology to change to a hypothesis-
driven research field. This is the perfect role in which HPC 
can be used. However, computational biology in my estimation is 
today where computational physics was in the late 1960s and 
1970s. The use of HPC will be required if biologists are not to 
be overwhelmed by the amount of data being generated. This is a 
particular feature of biology. The sequence of the human genome 
has been completed. Another 100 genomes have been sequenced and 
published and another 700 are in the works. The amount of 
information that can be derived from these genomes is immense.
    In addition, experimental methods and chip designs have 
increased the enormity of the data. For example, we now have a 
chip about the size of a microscope slide that is able to 
detect every known mammalian and bird virus. This is how SARS 
was detected. These chips have the potential for producing 
hundreds of thousands of data points and even millions of data 
points for altered disease and up-regulation, down-regulation, 
and appearance of proteins in cancer.
    In addition, the National Cancer Institute has invested in 
other experimental techniques, such as mass spectroscopy, 
imaging, nanotechnology, et cetera, all of which will 
contribute more data. The data must be annotated, curated, and 
analyzed in order to extract information, but it is not just 
information that we are after; it is actually knowledge.
    The goal is not to create just databases, but the goal is 
to create knowledge databases. These bases must be usable by 
nonexperts in computer science, especially by clinicians, if we 
are going to have translation into medical benefit. This is a 
challenge for HPC and the ability to store data.
    The ABCC has been in the forefront of the effort the expand 
the application of HPC into the field of molecular biology and 
cancer research. We provide advanced computational 
technologies, high-performance computer software, scientific 
expertise, and scientific support to both NIH internal 
researchers, NCI, at the universities and research institute 
scientists around the country.
    One example of this is in the discovery of cancer 
biomarkers, which is extremely important to the NCI because 
early detection is key for treatment of cancer. The ABCC has 
been involved with cancer researchers at Texas A&M, M.D. 
Anderson, Eastern Virginia, and others to analyze data for 
biomarkers for different cancers, and we have helped identify 
markers for bladder and colorectal cancers which could lead to 
inexpensive screens for cancer.
    The ABCC offers or provides training and related outreach 
programs to scientists in computational biology technology, and 
I am pleased to say that we have begun offering classes in 
molecular modeling and our classes are filled within 30 minutes 
of being announced.
    We also reach out to other agencies to adapt technology 
from other fields to molecular biology. For example, we adapted 
technology developed by the National Security Agency in 
cryptological pattern recognition to the study of, recognition 
and study of human genome work, especially tandem repeats. 
Tandem repeats are pieces of DNA bases that are expanded in 
many disease states. We were able to use technology developed 
by the National Security Agency and we could scan 150 million 
bases on a chromosome in 2 seconds and find all the tandem 
repeats, and we could do the entire human genome in 2 minutes, 
something that had never been done before.
    We have also applied field programmable gate arrays to 
certain biological algorithms to exceed thousandfold increases, 
but programming field programmable gate arrays is tremendously 
difficult.
    The ABCC has an ongoing funding line in the National Cancer 
Institute and the NIH budget, which enables our staff to 
provide the missionary work of enhancing cancer research 
protocols with the advanced tools of HPC. This is useful and I 
think a critical element in the fight to cure cancer. The 
Congress and the Administration, the NIH and the NCI, should be 
commended for developing policies and a program that fund these 
activities.
    I do have a couple of recommendations to make. As I 
mentioned before, cancer research and molecular biology 
scientists are substantially behind physical scientists in the 
application of HPC to their research procedures and protocols. 
Accordingly, the outreach program of ABCC and other institutes 
that promote HPC as a basic research tool in their biosciences 
should be encouraged and supported in their work. The National 
Cancer Center's program should embrace and promote the use of 
HPC tools and approaches as they credential and support 
programs of the external centers, such as Anderson, Sloan-
Kettering, San Francisco, and others. The center's program 
should also consider funding HPC applications at new emerging 
centers across the country, as opposed to the low level of 
support and sometimes the non-forethought of needing high-
performance computing for their analyses.
    I recommend that the Congress and the Administration 
encourage a fundamental change in the education of biologists 
in our university community to provide much greater emphasis on 
computational biology, which I believe will be the foundation 
of future biology and medical research. As a matter of fact, I 
just finished a RAND report to the National Cancer--National 
Intelligence Committee, on technology for the future, in which 
the statement is made that ``The intersection of high-
performance computing and biology and the fields that are 
developed through this effort, such as nanotechnology and 
biomedical materials, will be the future.''
    The third thing I recommend is that the Congress and the 
Administration raise funding for applications of HPC in 
biomedical research to that of an NSF program and that agency's 
track one and track two supercomputing program, and that a 
major flagship national center similar to the model of the 
National Center for Atmospheric Research in Boulder should be 
commissioned and funded on the main campus of the NIH to create 
a critical mass for computational medical research, with 
emphasis on cancer, heart disease, and other genetic diseases. 
This new center should possess a variety of computer 
architectures, should be added to the NIH Teragrid and the DOD 
research and engineering network to make its resources widely 
available for biological researchers around the country.
    Biology is a particularly complex and difficult field. It 
requires the synergy of physicists, computer scientists, 
mathematicians, biologists to all work together and to be able 
to speak the common language. One of the difficulties in 
translating HPC into action in biology is that the person who 
is performing the simulations and helping the biologists 
analyze their data must be able to understand the language that 
the biologists are speaking.
    I recommend that the ABCC outreach program that brings HPC 
tools and resources to cancer researchers across the country be 
extended and expanded, especially to new cancer centers that 
are coming online.
    That concludes my remarks and I appreciate the opportunity 
to appear before the Committee today and welcome questions 
regarding my testimony or the field of HPC applications in 
biomedical research.
    [The prepared statement of Dr. Burt follows:]

        Prepared Statement of Stanley K. Burt, Ph.D., Director, 
                  Advanced Biomedical Computing Center

    Over the past couple of decades many important milestones in 
biology have been obtained. These include completing the genomic 
sequences from several mammalian genomes, including human, and 
producing draft sequences for several additional genomes. Also, new 
technologies allowing for simultaneous measurement of mRNA expression 
levels for thousands of transcripts and application of this method to 
RNA samples from tumors and normal tissues have identified many genes 
whose expression is influenced by cancer and other disease states. 
Further improvements in this technology and other discoveries have led 
to chip technologies capable of simultaneously monitoring the entire 
genome for person-to-person variations including both single nucleotide 
polymorphisms (SNPs) and other larger alterations such as deletions and 
duplications. Other data derived from experiments that measure microRNA 
expression levels and transcription factor binding studies have also 
contributed to the extensions of this technology. Finally, measurements 
of protein expression levels using high throughput mass-
spectrophotometry and chip-based tissue and antibody arrays have given 
biologists the ability to correlate changes in mRNA expression with 
changes in protein expression levels that may contribute to the disease 
process or at least be markers for these altered physiological states.
    New instrumentation in microscopy has allowed for simultaneous 
monitoring of cells responses to drugs or other agents in parallel. 
Further, con-focal imaging techniques have allowed for multiple slices 
of the same fields to be examined in detail so that a three dimensional 
image of a specimen can be reconstructed. Whole-animal imaging is also 
being used to study drug distribution throughout an animal's tissue. 
Other methodologies allowing for higher levels of protein expression 
and purification are being leveraged to allow for more direct 
biochemical metrics of an enzymes function to be collected. High 
throughput binding technologies can be used to determine affinities of 
proteins with cofactors and drugs. Better docking software applications 
now exist to screen some of these interactions in-silico. Newer, more 
sensitive and reliable methods have been developed out to identify 
protein-protein interactions. The biomedical literature has also grown 
dramatically as all of these new methods and the data associated with 
each of them has increased.
    Taken together, these new methods and the need to process and 
analyze the data produced by them have resulted an explosion in the 
need for high-performance computing in biology and medicine. This need 
requires both increased capacity, as the sheer volume of data generated 
is considerable, and also increased capability. One of the confounding 
problems associated with the needs analysis of this problem is that 
there does not appear to be any single solution to the problem. Because 
of the diversity in the algorithmic requirements for analysis of each 
of these data types, no particular computer hardware seems suited for 
all of the problems. Thus, some of the problems are embarrassingly 
parallel, meaning they are ideally suited to a cluster environment. 
Good examples of this type of application would be comparing fragments 
of one genome to another, where each computation is entirely 
independent of the other. Advances in microarray plating technology now 
allows for increased spot density. This translates into a tremendous 
increase in the amount of data from a single experiment, at a 
significantly reduced cost. In addition, since these experiments only 
produce useful results when they are run for many samples (e.g., tumor 
and normal tissues) a greater volume of data is produced. This is 
leading to the need for the biologist to have access to computers with 
more memory and higher processor speeds to allow the data to be 
analyzed in a reasonable time. Already, the ABCC has received requests 
from cancer biologists for help with genomic analysis of promoters, 
control regions, miRNAs, better annotation of the genome and comparison 
of genomes, understanding of fragile sites, sites of chromosome 
translocation, and the relationship to cancer of segmental 
duplications. In addition, the new 500K SNP chips are flooding 
researchers with data that requires big computers to process, store, 
and interpret. Cancer biologists want new methods to look at the data 
and estimate haplotypes and look for interaction among many loci. This 
and all of the abovementioned challenges require the use of HPC 
resources.
    Another area in which cluster computing can be useful is in 
biomarker discovery. Aside from prevention, diagnostic tools to detect 
cancer at an early stage are of great benefit to patients. Great 
efforts are being made to identify biomarkers from gene or protein 
expression profiles. One tool being used to find biomarkers is mass 
spectroscopy, which can identify proteins and their fragments based on 
their size and electrical charge. In mass spectroscopy experiments 
thousands of spectral peaks are produced. These peaks are then used to 
find biomarkers for proteins. Because there are so many data points 
that are trying to be fit to few markers, this can lead to false 
results because the problem is over-determined. In order to avoid this 
mistake, one needs to perform thousands of calculations to develop a 
consistent set of models to find the proper biomarker. The ABCC does 
this by using methods that converge on a model that has the same 
biomarkers in each solution, thereby guaranteeing a biologically 
relevant answer. This procedure can benefit from hundreds of 
processors, but large memory is not needed since each calculation is 
independent of the others. The ABCC has been successful in finding 
biomarkers for bladder cancer and colorectal cancer. Hopefully, these 
markers, which are derived from urine and human serum, will translate 
into efficient, inexpensive screens that can be used for early 
detection of these cancers.
    Another problem that confronts biological computing and cancer 
research in particular is the sheer volume of data that must be 
collected, analyzed and compared. Data already exists in older 
databases in many places and in different formats. Part of the problem 
is already being approached by the NCI through its caBIG (Cancer 
Bioinformatics Grid) initiatives of NCICB and it involves identifying 
and leveraging information technologies that facilitate data 
interconnectivity, amongst other goals. In this regard, the development 
and enforcement of data exchange standards through caDSR and caCore are 
designed to bridge the gap between a clinicians and a bioinformaticists 
perspective of a set of genomic data. In order to analyze and house 
this data, there needs to be a computational infrastructure and 
visualization capabilities. Furthermore, while distributed databases 
are convenient for data maintenance, the National Security Agency has 
found that having all the data reside locally, where it can be called 
into computer memory, is essential for rapid data scanning. This will 
require HPC resources with large memory resources. Also, database 
consolidation is not enough. There needs to be development of methods 
for the construction of a knowledge base in which nonexperts, 
especially clinicians, can query data from various sources. This will 
require a serious research effort in knowledge base development area, 
although some manufactures have obtained preliminary results in this 
area. In addition, because there are problems suited for both hardware 
configurations mentioned above, the data I/O infrastructure must also 
be able to be connected to both of these scenarios. Again, because of 
bandwidth issues resulting from the sheer volume of the data, this 
results in a need for new technologies in computer architectures.
    Another complicating factor in data combination and analysis for 
biological research is that while massive storage and bandwidth have 
become relatively cheap and abundant, the data can not only be from 
different sources but it can represent experiments in different scales, 
from years to femtoseconds--time-scales that go across orders of 
magnitude. This is a problem that is referred to as multiscale 
modeling, and it is a profound problem in computational science. 
Solving this problem will require a commitment of resources to advanced 
architecture development, more efficient algorithms, and clever data 
reduction.
    I will now address some computational bottlenecks for a few areas 
that the National Cancer Institute has identified for their roadmap.

Nanotechnology
    Nanoparticles typically have dimensions smaller than 100 
nanometers, which are smaller than human cells. Nanometer devices 
smaller than 50 nanometers can easily enter most cells. Nanoscale 
devices can interact with biomolecules on both the cell surface and 
within the cells. Despite their small size, nanoscale devices can also 
hold tens of thousands of small molecules such as a contrast agent or a 
multi-component diagnostic system capable of assaying a cell's 
metabolic state. This can provide a mechanism for detecting cancer at 
its earliest stages. Nanoscale constructs, such as dendrimers and 
liposomes, can provide customizable drug delivery to targeted cancer 
cells or tissues. This has already been demonstrated experimentally.
    While nanoparticles have great promise, it also has to be 
demonstrated that they are not toxic to normal tissue. The ABCC is 
supporting the NCI's Nanoparticle Characterization Laboratory through 
modeling of bulk properties and calculation of atomistic properties. At 
the nanoscale, the physical, chemical, and biological properties of 
matter differ fundamentally and often unexpectedly from those of 
corresponding bulk material because of the quantum mechanical 
properties of atomic interactions which are influenced by material 
variations on the nanometer scale. Modeling of bulk properties such as 
surface charge or shape is not difficult. The calculation of atomic 
level quantities is a huge computational issue, even atomistic 
calculations on quantum dots are beyond our current capability, and 
will require large increases in HPC.

Drug Design
    Over the years there has been great success in drug design using 
HPC. Drug design is usually done against a protein target, such as an 
enzyme whose function one wants to inhibit. A great recent example is 
the discovery of Gleevec, an inhibitor of protein kinase activity, 
which brings about complete and sustained remission in nearly all 
patients in the early stages of chronic myeloid leukemia. If the 
structure of the protein is known, docking calculations can be 
performed. This usually involves docking thousands of molecules into an 
active site and scoring the resultant interaction. If the docking is 
done with rigid molecules, the calculations are fairly trivial. If, 
however, flexibility is allowed, and most proteins and ligands do flex, 
then the problem becomes enormously computationally expensive.
    If the protein structure is not known, and the protein is not 
similar to another one, then one must perform ab initio structure 
determination. David Baker's group at Illinois took approximately 150 
CPU days to determine the structure of the CASP6 target T0281. Also to 
do a docking interaction between two proteins took 15 CPU days. He 
makes particular note that his group is limited by computational power. 
Our group has studied the enzyme mechanism of many enzymes involved in 
cancer. For an enzyme named Ras, which is mutated in over 30 percent of 
known cancers, we modeled 1,622 atoms of the protein by molecular 
mechanics and only 43 atoms by quantum chemistry. These studies took 
several years and were bound by computational power. To calculate 
reaction surfaces normally takes several months of time on HPCs. 
Luthey-Schulten's group at Illinois did molecular dynamics simulations 
of Imidazole Glycerol Phosphate Synthase, an enzyme involved in making 
DNA and RNA. It took 10 hours, 12 hours, and 40 hours to animate one 
nanosecond on three cluster machines (with different processor speeds). 
It takes many nanoseconds of simulation to just relax the systems to 
prepare for further simulations. It has been estimated that to go from 
nanoseconds to milliseconds will require an increase in computer 
capacity of approximately 1,000,000. This can only be achieved by the 
combination of improved hardware and software.

Integrative Biology
    Computer aided design of HIV protease inhibitors remains one of the 
most successful stories in modern biology. Although this was a 
remarkable achievement, the complexity of a single viral particle pales 
in comparison to characterizing the complete catalog of the cell (the 
proteome) and the full map of the interactions of the members of the 
proteome. For a subset of interactions of the proteome, the immunome, 
the combinatorial problem of treating all possible pairs in the 
immunome (1,000,000 of them) escapes the capacity of current computers.

Synzymes
    There is great interest both in academia and industry for the 
creation of artificial enzymes that are much smaller but duplicate the 
enzymatic activity of the large natural ones. Because they are smaller, 
they can be tethered to other molecules or nanopartices, such as 
dendrimers or liposomes, and delivered to a particular targeted area 
such as a tumor cell. The ABCC staff has experience in this area. We 
modeled a particular inorganic catalyst know as Mn-salen, which is used 
commercially in the chemical industry for epoxidation reactions. After 
studying this reaction, we were able to convert this catalyst into one 
having biological activity and could act as a free radical scavenger. 
This could be useful for traumatic injuries, strokes, or even for 
cancer. However, this falls into the same category as enzyme mechanism 
studies and the calculations take months and months to perform. 
Complete characterization of these reactions took several years running 
on fast HPCs.

Specialized Hardware
    Being able to take advantage of specialized HPC resources and 
software written for those resources can lead to dramatic increases in 
time to solution. In one instance, the ABCC staff in a research 
parternership with several NCI biologists investigated how to rapidly 
scan for microsatellites (tandem repeats). Tandem repeats are groups of 
DNA nucleotides ranging from two to sixteen bases that are expanded in 
several diseases. For example, in normal people there is a pattern of 
DNA nucleotides, CAG that is expanded 10-35 times. In Huntington's this 
same pattern is expanded between 36-121 times. In the past finding 
these repeats were found in a heuristic and probabilistic manner on 
conventional computers.
    Using specialized hardware such as bit matrix multiply and pop 
count, which had been requested by the NSA to be incorporated in the 
machines they were using in order to perform rapid pattern matching, 
we, along with industrial programmers, were able to drastically reduce 
the time to find all tandem repeats on chromosomes and the entire human 
genome. To scan a chromosome of approximately 150 million bases took 2 
seconds. To scan the entire human genome took 2 minutes. We discovered 
47 potential disease sites, 8 of which could be associated with cancer, 
and we more than doubled the known numbers of repeats. We also used 
this specialized hardware search for another genomic feature, known as 
segmental duplications, which are associated with diseases. This 
involved finding clusters of DNA bases approximately tens of thousand 
bases long that are separated by approximately 1 million bases from 
another cluster of bases that are the complimentary complement of the 
original DNA base cluster. When these complimentary clusters find each 
other during replication they combine and huge sections of the genome 
are excised. We could not have done this without these specialized 
hardware features.
    We have also used FPGAs, which are reprogrammable hardware and 
support the custom computing needs that are characteristic of data-
intensive problems. We programmed the FPGA for a powerful sequence 
alignment algorithm known as Smith-Waterman. The Smith-Waterman 
alignment method is a powerful algorithm for aligning sequences in 
which there may be gaps and one is trying to find the ``best'' 
alignment. This algorithm is widely used in the biological community 
but is particularly computationally demanding. We obtained speed-ups of 
over a thousand fold. However, the difficulty is that the programming 
of FPGAs is not a trivial task, and one that would not be normally 
within the expertise of a biologist. However, FPGAs offer great promise 
because there are expected to be huge increases in performance on these 
types of machines.

Recommendations
    It has been said that biology will be the science of the 21st 
century. Due to the complexity of biology, the sheer volume of data, 
the fact that the environment of a cell, (particularly for cancerous 
cells) must be taken into account means that biology must be tackled 
using a systems biology approach. This means that teams of scientists 
such as biologists, computer scientists, mathematicians, physicists, 
and chemists should work on these problems in conjunction. In order to 
do this, it will require cross-training to have a meaningful dialogue. 
I believe that in order for the United States to remain competitive we 
should devote funding to education and training in the above 
disciplines. We also need to find mechanisms to encourage young people 
to enter the scientific field. I have seen for several years the lack 
of U.S. citizens applying for jobs in the ABCC. I believe that this 
reflects the national trend.
    As biology matures the use of HPC in biological research will grow. 
There is clearly a need for large memory assets coupled with fast 
processors. I believe that cluster computing will still have its place, 
but as the problems grow in size and complexity, the need for HPC 
resources will be inevitable. One can already see this trend in Europe 
where several national centers have made purchases of Blue Gene 
machines, and others have made investments in large memory machines.
    There needs to be funding of new computer architectures, 
specialized hardware, faster interconnects, etc. One area of funding 
that is especially important is software development. One thing holding 
back HPC development is that the software available today is not 
written for HPC machines. Sometimes software engineers spend 
considerable time to port nonparallel applications to parallel machines 
without much increase in speed or efficiency. We are running ``old'' 
software on newer architectures. Along with developing new software, 
research into new compilers must be encouraged.
    I also recommend that the United States fund several centers for 
Integrative Computational Technology for Systems Biology. These centers 
would provide for the integration of biology with strong computational 
infrastructures and analytic tools. These centers need to provide 
intuitive, visual interfaces for biologists with real-time interactive 
data analysis. These centers could also serve as training facilities 
and facilitate communications between scientists of diverse 
backgrounds, disciplines, and expertise within a common framework. 
These centers would also facilitate the interplay between discovery and 
hypothesis-driven science. Several other countries are already creating 
such centers.
    Maintaining a leadership role is vital for the economic health of 
the United States. We need to maintain our leadership in HPC in order 
to have the advantage in intellectual property, which is connected to 
our economic well-being. Support for our HPC industry is vital. 
Countries such as Japan, China, and India are making substantial 
investments in HPC. We need to do the same.
    The need for supporting HPC extends across all of the hard science 
disciplines. I hope that I have been able to show in this statement 
that the increased need for this support is arising from biology. A 
recent RAND report entitled ``The Global Technology Revolution'' was 
prepared for the National Intelligence Council. In this report it 
summarizes how the future will be determined by the intersection of IT 
and biology, and the industries such as nanomaterial, materials, and 
biotechnology that are spun from this intersection. Clearly, the future 
is in this area. We should make the investment now.

    Senator Ensign. I want to thank all of you for some very 
fascinating testimony.
    I am going to take about 5 minutes and cover one topic and 
then I am going to turn it over because Senator Cantwell has to 
leave and let her spend some time, and then I am going to come 
back and ask a few other questions.
    I want to explore this because at this hearing we have 
academia represented, we have the private sector and industry 
represented, and I have always believed that it is a 
fundamental role of government to conduct basic research. 
Applied research is more the responsibility of the private 
sector, and sometimes there is that little nebulous area in 
between where sometimes we use government programs to try to 
bridge the gap. But when we are looking at the idea of basic 
research and funding basic research in the area of HPC--now I 
am using that acronym; I did not know the acronym before we got 
it today, but I will join right in with it.
    Can you take a minute each and talk about where the line 
between basic and applied research is when it comes to HPC?
    Dr. Wladawsky-Berger. If I may start, in the past the bulk 
of the basic research was in the actual technology component. 
While some of that, especially pushing the leading-edge, might 
be needed, what we are finding is that the really complicated 
problem is how do you put the whole system together, including 
software and applications. That goes well beyond the components 
and that is where the testbeds, the pilots focusing on 
applications, are so important.
    For example, what Dr. Burt talked about. Focusing a problem 
on advanced cancer research or brain mapping research, which is 
a major area, brings all these ingredients together. The 
testbeds are extremely important that become a kind of boundary 
between researching the components and beginning to get it into 
the marketplace, and then the private sector takes that and 
then they themselves bring it to a lot more applications, a lot 
more business areas, and make it less expensive and so on.
    Senator Ensign. Thank you.
    Mr. Jehn.
    Mr. Jehn. Yes. I would like to just echo what you said a 
moment ago, and that is to say that this is really a classic 
example of a place where Federal support for R&D is justified, 
and the government itself has recognized this in the reports I 
have cited. I would simply echo those as well, that in addition 
to basic research that would develop the building blocks that 
companies like IBM and Cray could use to develop next-
generation supercomputers, support for research is also 
required in the area of basic architectures and ultimately to 
build prototypes.
    In fact, the Federal plan for high-end computing and the 
similar report that the Defense Department released the so-
called IHEC report, proposed a four-component R&D program: one, 
basic and applied research; two, applied developmental work; 
three, building prototypes; and four, establishment of a 
handful of laboratories in the government that would 
consolidate this. A program like that should support R&D in 
government, in industry, and in academia. That systematic 
approach is what is missing today in Federal policy.
    Senator Ensign. Mr. Waters.
    Mr. Waters. When I think of basic research, I think a 
little bit back in the past when it all used to sort of happen 
in one place, whether it was a lab or a campus environment. I 
actually came from Dr. Burt's lab. About 20 years ago, it was 
my first job out of school. What happened was that the basic 
research occurred on one campus.
    I think what is occurring is that basic research is getting 
so expensive in some examples that the instruments themselves 
cannot be duplicated. So that the need is for a distributed 
architecture that we have not had for basic research, and I 
think that is where a lot of emphasis needs to be put, whether 
it is grid computing, whether it is distributed storage, the 
middleware or the software that sits and controls how people 
use distributed computing. That is not an area where I think we 
have focused a whole lot.
    Senator Ensign. Is that part of the government role or is 
that more the role of the private sector?
    Mr. Waters. I think the government does play a role and can 
play a role in that basic research, because it is not an area 
that I think industry has focused on.
    Senator Ensign. Mr. Lombardo?
    Mr. Lombardo. From my perspective, the basic research 
really chimes in with the Grand Challenge problems. We are 
looking at answers and solutions that may be a decade away, 
maybe 20 years away. So it fits very nicely in that realm, and 
I can see where the Federal Government would want to fund that 
research. And the opposite is also true. If there is a solution 
that is ready in 3 months, 6 months, 18 months, that may be 
better left for the corporate side or the corporate sector to 
handle.
    Senator Ensign. Thank you.
    Mr. Garrett?
    Mr. Garrett. Sir, from a user perspective, most of our 
research and development is going into the tools that we use to 
do the analysis, not into the high-performance computing area, 
though our tools are running neck-in-neck with the high-
performance computing as far as which is getting further ahead. 
So that is part of it.
    Just as an example, the tools that we used 15 years ago 
which were using high-performance computing are today run on 
laptops. And we still use those tools, but now they have 
migrated down to where the user can use them on a laptop. So we 
develop the capacity and as the capacity comes online we need 
it, but there are cases where we need more.
    Senator Ensign. Does Boeing, as a private company needing 
HPC, use or have access to supercomputers at an academic 
institution?
    Mr. Garrett. We do as needed. We have our own, for example, 
Cray X-1 in our Bellevue campus, which we do use, which is 
supplied by Cray, obviously. And it has been upgraded 
tremendously in the last 5 years. But we also have access to 
other companies' usage, as well as sometimes, based on 
government contracts, to government computers to be used on the 
defense side. We do not use those very much on the commercial 
side, but in support of defense contracts.
    Senator Ensign. Dr. Burt, Please hold off on your answer 
until after Senator Cantwell is finished with her questions, 
because I have some even more detailed questions for you. Being 
a veterinarian, I am very interested in how high-performance 
computing can impact medical treatments as well. Now I will 
turn it over to Senator Cantwell.
    Senator Cantwell. Thank you, Mr. Chairman, and thank you 
for your indulgence.
    I want to follow-up on where the Chairman went as far as 
the next phase of supercomputing and what we need to do as it 
relates to the House legislation and Senate legislation that we 
have been considering. Mr. Jehn, you talked about this stage of 
adaptive computing so that you can basically--just the diverse 
processing and scientific requirements. Mr. Waters kind of 
alluded to it in the sense of networks with a thousand times 
capabilities recommendation.
    So what is it specifically that we need to do in that area? 
And I know that the Chairman just asked in the sense of what 
role we should play. But are you saying that this is an area 
that is the next phase and that we ought to be more specific in 
outlining this as far as our competitiveness?
    Mr. Jehn. Well, a government-sponsored R&D program would be 
much more specific and systematic, than the government has 
funded to date. However, I again go back to the report of the 
High-End Computing Revitalization Task Force of 2 years ago. 
They laid out a roadmap, a technological roadmap, that would be 
a candidate for a systematic R&D program. The government over 
the last 3 or 4 years has made enormous strides in this 
direction.
    As you may remember, I joined Cray a little less than 5 
years ago and was relatively new to this industry and this set 
of issues, and particularly Federal policy in this area. 
Frankly, there has been a very significant change in Federal 
policy and activity over the last 3 or 4 years that Dr. Szykman 
referred to as well. I think building on that momentum is what 
is required right now. At the moment individual agencies like 
DARPA and the National Security Agency are supporting R&D in 
specific areas that they feel are most appropriate and most 
applicable to their requirements, but we need similar energy 
and direction elsewhere throughout the Federal Government. And 
we need better coordination among the various agencies to 
ensure that nothing slips through the cracks and that we are 
not duplicating things unnecessarily.
    But there is plenty of room for everybody to contribute and 
I would advocate them doing so.
    Senator Cantwell. Well, when you talk about systematic R&D 
and Mr. Waters talks about the implications of TCPIP having 
been developed by DARPA, it sounds like we are talking about a 
similar focus to really bring together the great computing 
science, but with an easier use for the individual researchers 
in that application; is that correct?
    Mr. Jehn. Well, certainly in the field of high-performance 
computing there is no mistaking that as capable and powerful as 
these systems are, they are very, very difficult to use. 
Imagine programming a system that contains 100,000 laptops, 
basically, 100,000 microprocessors. It is a daunting challenge, 
and the focus of the DARPA program has been and is to develop 
systems that are far easier to use, more accessible to a wider 
range of users, and I think this is a trend, a bit of momentum, 
again to use the same word, that we need to build on. I think 
Mr. Waters' example of the Internet is a great example or a 
great analogy here, where a very narrow, focused bit of 
technology was expanded to----
    Senator Cantwell. Explosive.
    Mr. Jehn. Yes, it has been explosive, with wide 
availability and applicability to enormously different areas. 
That is what we need to promote in this area as well.
    Senator Cantwell. Mr. Garrett, I think that when you were 
using the aviation applications that you were talking about, I 
do not know if that is what Mr. Lombardo would talk about, the 
midrange applications, but it certainly has great applications 
for your competitiveness today.
    Mr. Garrett. Yes.
    Senator Cantwell. What are your competitors doing in this 
area?
    Mr. Garrett. Our competitors, Airbus specifically, is doing 
the same thing. They are investing, making those same 
investments.
    Senator Cantwell. Where do they get their research?
    Mr. Garrett. Well, they get that either through themselves 
or academia or through the government.
    Senator Ensign. If you could elaborate on that point, 
because we talked about the U.S. versus other nations with Dr. 
Szykman on the first panel, and Airbus is obviously your 
competitor. On the applications side, do we have an advantage? 
Does the farther we stay ahead in the United States on high-
performance computing give Boeing an advantage in its 
competition with Airbus?
    Mr. Garrett. Yes. In the long run, we have had the issues 
of our competitiveness and what has played out in the 
marketplace has been of benefit to us in the last year 
especially. At Farnborough that is playing out. But the bottom 
line is the performance. Our ability to meet our performance 
commitments is getting more and more attention because of the 
lack of that in many campaigns with our competition.
    That does not happen by accident. It happens because of our 
ability to predict the airplane's capability from a weight, 
engine, and drag standpoint has been significantly improved and 
actually has been our competitive advantage for years. It is 
just coming out now in some of these key campaigns. As they get 
more and more aggressive, this is where we start to shine, in 
our ability to use these tools.
    It is HPC is one element, but it is the step ahead that we 
have in our proprietary analysis tools and the development of 
those computational methods which is reducing the risk. It is 
basically risk reduction and our ability to hit our target 
where we think we are going to be, and our customers are 
acknowledging that and we are getting credit for that in the 
marketplace now, which is nice to see.
    Senator Cantwell. Thank you, Mr. Chairman. I will leave you 
to query Mr. Burt on a lot of areas. But I certainly agree with 
his philosophy that the intersection of supercomputing and 
biology is where the future is, and we would like to think a 
piece of that is in Seattle. So thank you.
    Dr. Burt. Senator Cantwell, if I may just make one comment. 
When I mentioned data-intensive computing, I have been for 
several years on the scientific advisory board for Pacific 
Northwest National Laboratories in their computer science 
initiative. One of their major concerns is how do you do this 
data-intensive computing and how do you gather it, because they 
of course are concerned across many fields, not only biology 
but national security, homeland security, et cetera, et cetera. 
So it is a constant. So I serve on a panel that is devoted to 
looking at how they are approaching this problem.
    Senator Cantwell. Thank you.
    Thank you, Mr. Chairman.
    Senator Ensign. Thank you.
    Let me start with Dr. Burt. You mention that because of the 
complexity of biology and the life sciences, a cell is 
infinitely more complex than an airplane. It just is.
    Dr. Burt. Yes, he is only doing airplanes.
    [Laughter.]
    Senator Ensign. So it would seem to me that 
supercomputing--and I remember when I was in veterinary school 
in the 1980s, the talk about pharmacology and the idea of some 
day replacing animal research with computer modeling. With a 
lot of the research, without high-performance computers you are 
not going to even come close to doing some of the things that 
Mr. Garrett is talking about in predictability. I think you 
called it hypothesis-driven.
    Dr. Burt. Hypothesis-driven. It means that you can use the 
computer to help you generate various models, which will then 
help direct where the experiments need to be run.
    Senator Ensign. Right, and that would seem to me, similar 
to the airplane example, how it shortened the life-cycle down 
and the development cycle of new products, the same thing with 
new drugs, the same thing with new treatments, the same thing 
with a lot of different things in the field of medicine.
    The question is on the Federal Government's role. We 
doubled the budget for NIH and it continues to go up from 
there. One of the goals that we have is to develop the budget 
for the physical sciences in the same way. Do we need to do 
something differently up here? Since we doubled the budget, we 
hate to get into telling the scientists what to do with their 
money. It has worked fairly well with Congress not politicizing 
that too much. But are there other areas that need focus or 
should we at least say we need, as a national priority to 
invest for the life sciences into high-performance computing? 
Do we need to start directing NIH and some of their funding to 
do that?
    Dr. Burt. I believe so, Senator. The ABCC happens to be 
probably the only example of a truly integrated high-
performance computing center of the NIH involved in biomedical 
research. The Center for Research Resources has made grants to 
other universities for computing and et cetera and I think they 
have been successful in that. But I think that we are just now 
beginning to realize that you really need, because you have to 
take a systems approach to biology because you must take into 
account not only the cell, not only the proteins, but the 
pathways, and especially in cancer it has been shown that the 
environment of the cell plays a large role.
    So we really need to fund something like I proposed in 
order that we can get more people from different disciplines 
involved. As I said, one of the things that makes the--I did 
not say this, but one of the things that makes the ABCC 
successful is that the bioinformaticists in my group all have 
experience in the labs as well as with computer scientists. The 
people who do the modeling for me also are physicists and 
quantum chemists and mathematicians who have now received 
enough experience that we can translate. So that is a real key 
to applying HPC in biology, is the synergy, this collaboration, 
and we need cross-training so that people can speak the 
languages.
    Senator Ensign. Well, I would like to work with you, if you 
could make yourself available to work with my staff, and try to 
come up with some ideas along these lines. I do not like to 
just have hearings. I like hearings to lead to actually policy. 
So I would like to work with you on some of the things that we 
are talking about today.
    Sorry, doctor. You wanted to comment?
    Dr. Wladawsky-Berger. Yes, if I may add. Dr. Burt said 
something really, really important, which is more and more we 
need to take a system approach to these problems. If I could 
use that to link back to the innovation authorization bill, 
because in the past the bulk of the fundamental research was 
components in the laboratory, and then we threw it over the 
fence and then people built things with them. The problem now 
is that the things we are talking about building are systems of 
such extraordinary complexity that we need to actually do the 
research in how do you build those systems effectively. And 
since those systems have to be usable by human beings--
otherwise why build them--how do you, for example, visualize 
the results so that the physicians or the veterinarians or 
whoever is using them can actually work with them?
    So the change that has happened is it is not just in the 
laboratory; it is almost more, there is a lot of marketplace 
innovation, if I may use that term, in how do we build these 
systems and make them usable by human beings?
    The Internet, by the way, is probably exhibit A of what we 
are talking about, which is we did not just do the TCPIP in the 
lab. NSFNet actually built the Internet, and then the World 
Wide Web came out of that. So that is something that is very 
different that links HPC to the innovation bills that you have 
been working so hard to authorize.
    Senator Ensign. Mr. Waters, along those lines, because you 
folks are more involved obviously with networks than anybody 
else here--you talked about the Internet2. If we are linking 
Mr. Lombardo with other places around the country, what are the 
requirements? We are talking about more broadband in the 
country. Are we just talking fiber? Are we talking compression 
technology? What is necessary for that Internet2--I would 
imagine when you are talking teraflops--or what is the next one 
beyond teraflops?
    Dr. Wladawsky-Berger. Petaflops.
    Senator Ensign. Petaflops.
    Dr. Wladawsky-Berger. Ten with 15 zeroes.
    Senator Ensign. Right, I knew the 1015. I just 
wanted to confirm the name.
    When we are talking about transferring that type of data, I 
would imagine that over a traditional phone line the transfer 
might be a little slow. I am just guessing. But what kind of 
network capabilities are required? And on Internet2, how long 
into the future before Internet2 is available?
    Mr. Waters. The networking requirements that I believe are 
coming out of the HPC environment are measured in the hundreds 
of gigabits per second. So in your home line you may have a meg 
and a half. A gigabit is 109, so you can kind of get 
a feel for the orders of magnitude that we are talking about.
    That is just the start. That is the initial Internet2 
infrastructure at 100 gigabits per second. We believe that that 
will grow to four or five times over the next couple of years 
as research needs are----
    Senator Ensign. What is the physical----
    Mr. Waters. The physical is the infrastructure that has 
been invested over the last decade. The fiber itself is 
frankly, the physical fiber, is probably already there to link-
up the institutes where important. It is really the electronics 
that have to be put on the ends of the fiber, and that is where 
some investment needs to be made; and also the control systems, 
the things that allow a researcher in San Diego to have 100 
gigabits at a particular time of day and then perhaps a 
researcher in Pittsburgh, being able to switch that bandwidth 
to that researcher in Pittsburgh, because you cannot 
necessarily provide bandwidth everywhere at the peak capacity 
at all times, and we want to be able to shift that bandwidth to 
individual users and applications over time.
    Senator Ensign. Mr. Lombardo, could you just make a comment 
on what that would do to centers like your own, being able with 
your researchers to communicate with other researchers, not 
just working on your computer there but actually, just like you 
join PCs together, you can actually collaborate a lot more?
    Mr. Lombardo. That is exactly right. In fact, for us, we 
jumped onto the Internet2 backbone about 6, 7 years ago, and 
the amount of data that we can currently download is 
incredible. Having access to such data enhances the research 
outcomes in modeling and simulation problems. We are currently 
in a planning stage to develop a center for the simulation and 
modeling of brain disorders, such as schizophrenia and 
Alzheimer's. But what is key to this research--and this is one 
of the rules I always like to mention about supercomputing: we 
need computers, storage and, high-performance networking 
capability. Without which we would not have the ability to move 
these enormous amounts of data without waiting--in some cases 
you would wait weeks and on Internet2 it may take you an hour.
    Internet2 access is both critical and an essential 
component of HPC. And the fourth element of HPC, which I still 
think is the most important, is the access to very smart 
people.
    Dr. Burt. Senator, I would like to just speak to that just 
very briefly. Since you are a veterinarian, there has been a 
big emphasis in Frederick on doing animal imaging. Now the goal 
is to be able to share those images with the people in 
Bethesda. So we have had to put in a big pipe so that people 
can stand in what we now have is a wall, so that you can look 
at this thing on the big screen and in three dimensions, but at 
the same time it can actually be seen by pathologists and other 
people like that. We have them at Frederick, too. But other 
people at the NCI in Bethesda, a distance of only 30, 35 miles, 
but we need to have it to where you can see it; as it moves, 
each person can see it move in real-time.
    Senator Ensign. Well, thank you all. We have rules in the 
Senate about the length of Subcommittee hearings, and I think 
they did that just because Senators' attention spans are often 
not that long. But I thank all of you. It has been a 
fascinating hearing today, and once again compliments to my 
colleague for inspiring this hearing.
    The hearing is adjourned.
    [Whereupon, at 12:35 p.m., the hearing was adjourned.]


                            A P P E N D I X

        Prepared Statement of Tom West, CEO, National LambdaRail

    Dear Mr. Chairman,
    On behalf of the member organizations of the National LambdaRail, 
we thank you for the opportunity to provide a prepared statement for 
the hearing record on high-performance computing.
    We applaud you and your distinguished colleagues for their 
dedication to sustaining and stimulating investments in technology for 
this Nation's innovation and competitiveness in the 21st century. One 
of the most effective ways to advance our Nation's research capacity to 
lead the world in innovation is ubiquitous access to high-performance 
computing resources (HPC). High-capacity optical networks are the means 
by which the scientific community broadly harnesses HPC resources for 
innovation and competitiveness.
    Today a new global network infrastructure owned and operated by the 
research and education (R&E) community has been deployed, and is being 
utilized. In the United States, National LambdaRail (NLR) owns a 
nationwide networking infrastructure that leverages regional and local 
efforts to provide a flexible infrastructure capable of supporting 
multiple, advanced research and education networks--a ``network of 
networks.'' NLR is available to all researchers in academe, Federal 
agency laboratories and non-profit and for-profit research 
organizations. It serves researchers in all scientific disciplines, 
providing the critical advanced network infrastructure to access the 
Nation's high-performance computing facilities for advancing big 
science initiatives. The NLR infrastructure is the result of over 3 
years of work and nearly $120 million in funding by its members.
    The mission of the NLR is to build an advanced, nationwide network 
infrastructure to support many types and levels of networks for 
research, clinical, and educational fields. This infrastructure 
consists of 15,000 miles of fiber and optical networking equipment, all 
of it owned by NLR. NLR's potential capacity is 40 10-gigabit (10 
billion bits per second) nationwide networks. By comparison, a 10-
gigabit network is roughly 10,000 times faster than today's commodity 
networks such as cable modem and DSL. The infrastructure supports both 
experimental and production networks, fosters networking research, 
promotes next-generation applications, and facilitates 
interconnectivity among regional and international high-performance 
research and education networks. Furthermore, NLR is scalable to 
accommodate the ever-increasing computing demands of the future.
    The hallmark of 21st century big science applications is multi-
disciplinary, multi-investigator research collaborations across time 
and space. This distributed approach can lead to more rapid and 
systematic solutions to society's most intractable challenges. High-
capacity optical networks are critical to leveraging innovation across 
these worldwide assets.
    Moreover, there is a growing urgency to develop new network 
technologies that scale to the growing needs of the worldwide R&E 
community and, later, to commodity Internet users. We are encouraged 
that the Administration's FY 2008 research and development (R&D) budget 
guidelines prioritize R&D in advanced networking technologies. NLR's 
high-performance network infrastructure enables the next generation of 
technologies, protocols, and services. This enabling infrastructure is 
critical to progress in essentially every interagency R&D priority for 
FY 2008--from homeland security to energy security, and from 
nanotechnology to complex biological systems and the environment.
    The focus on network researchers is a distinguishing characteristic 
of NLR. Fifty percent of NLR capacity is devoted to support network 
research projects at the forefront of developing and testing 
revolutionary, not just evolutionary, networking technologies and 
capabilities not possible in the laboratory or any other national-scale 
network.
    Undertaking this R&D requires an experimental testbed where network 
researchers can experiment with new approaches to all levels of 
networking technology. The results of this research will enable 
networks capable of supporting scientific projects in fields such as 
high-energy nuclear physics and radio astronomy, which require real-
time collaboration among scientists and manipulation of enormous data 
sets. Already, individual projects in these fields can usefully consume 
a majority of the largest network links available. Together, even a few 
of them could potentially overwhelm existing advanced research and 
education networks. And, these kind of bandwidth-hungry applications 
are spreading. Applications in almost every discipline are now emerging 
with the same need for big, broadband networks.
    While regional optical network (RON) infrastructure development 
emerged a few years ago, the formation of NLR spurred numerous new 
regional efforts. Now 14 NLR members operate 21 regional optical 
infrastructures to serve as the pillars of connectivity to NLR. Across 
the United States, an additional 15,000 miles of fiber-optic cable 
controlled by RONs significantly enhances access to rich high-
performance computing capabilities; unique, expensive research 
resources; and linkage of enormous amounts of data through federated 
databases.
    Importantly, NLR's diverse membership includes RONs as well as many 
of the Nation's premier research and education organizations, private 
sector technology corporations, and Federal agencies. Today, NLR's 
members include--

   Case Western Reserve University

   Cisco Systems

   Committee on Institutional Cooperation

   Cornell University/Northeast LambdaRail

   Corporation for Education Network Initiatives in California 
        (including the University and Community College System of 
        Nevada)

   Duke University (representing a coalition of North Carolina 
        universities)

   Florida LambdaRail

   Front Range GigaPop/University Corporation for Atmospheric 
        Research

   Internet2

   Lonestar Education and Research Network

   Louisiana Board of Regents

   Mid-Atlantic Terascale Partnership/the Virginia Tech 
        Foundation

   National Aeronautics and Space Administration

   Oak Ridge National Laboratory

   Oklahoma State Board of Regents

   Pacific Northwest Gigapop

   Pittsburgh Supercomputer Center/University of Pittsburgh

   Southeastern Universities Research Association

   Southern Light Rail

   University of New Mexico (on behalf of the State of New 
        Mexico)

    An excellent example of the computational environments enabled by 
NLR's infrastructure is the Extensible Terascale Facility (ETF) 
supported by the National Science Foundation. The ETF is a multi-
million dollar, multi-year effort that has built and deployed the 
TeraGrid, a world-class networking, computing and storage 
infrastructure designed to engage the science and engineering community 
to catalyze new discoveries. The Pittsburgh Supercomputing Center, one 
of the original TeraGrid participants, was the first organization to 
use NLR to connect its facilities to the nationwide TeraGrid facility. 
More recently, the Texas Advanced Computing Center acquired a 10 
Gigabit wave from NLR to connect Austin to Chicago. Oak Ridge National 
Laboratory is also using NLR for back-up waves between Atlanta and 
Chicago as part of ETF.
    Today more than ever, growth in our economy is increasingly linked 
to the investments made in fundamental research to advance computing 
and communications technologies. We urge your continued support for 
strengthening investments in America's future with a strong national 
research infrastructure for advancing discovery, innovation, and 
education.
    Thank you.
                                 ______
                                 
     Response to Written Question Submitted by Hon. John Ensign to 
                           Dr. Simon Szykman

    Question. How do other countries--particularly those in Asia--rate 
relative to the United States in the context of high-performance 
computing research?
    Answer. According to the most recent list of Top500 supercomputer 
sites * released in June, over 90 percent of the world's top 100 
fastest machines are located in the United States, Europe and Japan. 
Representation by other nations among the elite computing facilities is 
minimal. The locations of the top 100 machines break out geographically 
as follows:
---------------------------------------------------------------------------
    * See http://www.top500.org/.

        United States: 57 machines
        Europe: 18 machines

        Japan: 16 machines

        Canada: 3 machines

        Korea: 2 machines

        China: 2 machines

        Australia: 1 machine

        Russia: 1 machine

    As the most recent snapshot highlights, more than half of the 
world's top 100 machines are in the United States, about three times as 
many machines as Europe or Japan. Eighty-eight of the world's top 100 
machines, including more than two-thirds of those machines that reside 
outside of the U.S., were built by U.S. companies. Within Asia, the 
critical mass of supercomputing capability is clearly in Japan. In 
relative terms, representation by other parts of Asia is minimal; Korea 
and China each have two machines in the top 100, and India has none.
    The Top500 Supercomputers list provides a picture of investments in 
high-performance computing infrastructure. We can also consider related 
R&D investments as an indicator of how leadership may change with time. 
With the rapid growth of the economies of certain nations, it is 
important to consider not only a current snapshot, but to look forward 
as well.
    The European Union (EU) has taken a fundamentally different 
approach to high-performance computing than the U.S. has. The bulk of 
EU R&D investments in recent years has been through the EU's Framework 
Programme (FP), Both of the most recent Programmes (FP5 and FP6) have 
emphasized grid computing environments for high-performance computing. 
In this context, it is very important to note that grid infrastructure 
is not a substitute for tightly-coupled supercomputing architectures or 
centralized computing facilities.
    The U.S. Government's High End Computing Revitalization Task Force 
(HECRTF) recognized this limitation and deemed grid computing to be out 
of scope during the planning and roadmapping that resulted in the 
Federal Plan for High End Computing. The EU characterizes their grid 
computing investments as ``high-performance computing,'' but has no 
substantial R&D investments in high-end computing as characterized by 
the Federal Plan for High End Computing. Although the EU's Seventh 
Framework Programme is still in planning, current indications are that 
investments will continue expanding grid-based high-performance 
computing infrastructure rather than being aimed at R&D for new high-
end computing technologies.
    In the HECRTF context of high-end computing R&D (i.e., viewing grid 
computing technologies as out of scope), Japan is the only competitor 
to the U.S. in the global playing field. I mentioned in my testimony 
that 4 years ago the Japanese Earth Simulator System became the world's 
fastest supercomputer. It is now ranked tenth. Japan's 3rd Science and 
Technology Basic Plan (FY 2006-FY 2010), a Japanese national policy 
document, identified supercomputing as a key technology of national 
importance to the national infrastructure. Japan has announced plans 
for the development of a successor to the Earth Simulator System called 
the Next Generation Super Computer (NGSC). While the overall investment 
required for this effort is estimated at about $1 billion from 2006-
2012, to date the level of funding that has been approved is only about 
30 million in (Japanese) Fiscal Year 2006. The NGSC is expected to 
be a ten petaflops (10 quadrillion floating point operations per 
second) computing facility. The U.S. expects to have several petascale 
computing facilities by early next decade. As I pointed out in my 
testimony, global leadership should not be defined solely by the speed 
of the world's fastest machine. The NGSC will likely be a highly 
capable machine, but if the U.S. continues along its current trajectory 
of high-end computing R&D and infrastructure investments, the NGSC will 
not pose a threat to U.S. leadership in the high-end computing arena.
    At present, R&D capabilities and investments are significantly 
weaker elsewhere in Asia. Although China is emerging as an important 
global competitor as a producer of information technology (IT) 
equipment and products, China does not yet have a strong capability for 
innovation in the area of IT. Their core technologies and key equipment 
still rely on imports of technology and intellectual property. China is 
advancing their domestic IT innovation capabilities by focusing on 
integrated circuit and CPU technology and system and applications 
software, and not on high-performance computing technologies.
    There is some Chinese government support for high-performance 
computing, with around a dozen high-performance computing centers in 
China. However, these computing centers are not at the highest levels, 
as demonstrated by the fact that China has only two machines on the 
Top100 list (#35 and #53). Two Chinese IT companies (Lenovo and Dawning 
Group) have announced long-term objectives of building petaflop 
machines. but it is not clear how much R&D will go into supporting 
these efforts (as opposed to simply building machines through large 
investments technology that is within the state-of-the-art). In the 
long-term, China's domestic R&D capabilities are expected to improve 
and should be monitored. But at present and in the short- to medium-
term future, Chinese R&D in high-performance computing is not expected 
to be a threat to U.S. leadership.
    India has a strong IT sector, but one that is primarily limited to 
software development and IT services. India does not have a strong 
hardware sector, and is still working to establish a basic electronics 
industry. India does not yet have a world-class computer science R&D 
community. A national grid computing effort is only just beginning, and 
is still at a proof-of-concept stage. Physical infrastructure in India 
also lags behind that of China. India has pledged to increase R&D 
investments in the future, but given their current standing in the 
computing arena, they are not viewed as a significant threat at this 
time, and will need to make substantial strides in their ability to 
innovate from multiple perspectives (government commitment, workforce, 
infrastructure) before they are able to compete globally in the high-
performance computing arena.
    I hope that I have addressed your inquiry to your satisfaction. I 
would be pleased to respond to any additional questions you may have.
                                 ______
                                 
    Response to Written Questions Submitted by Hon. John Ensign to 
                      Dr. Irving Wladawsky-Berger

    Question 1. Please explain what features of present-day 
supercomputing do you feel will have the greatest impact on computers 
being used by the American public in the future?
    Answer. Hybrid programming models (such as the Los Alamos/DOE 
``Roadrunner'' which will use these software models to coordinate 
different chip architectures simultaneously) for supercomputers will 
become applicable to smaller systems--perhaps even PCs and game 
systems. This will greatly improve personal systems' performance and 
resulting capabilities--as Moore's law becomes less and less applicable 
over time. Chips can't just keep getting smaller to get faster (because 
each reduction in size brings a commensurate increase in heat) so new 
methods to combine multiple chips and varying chip architectures for 
even the smallest systems will be required. This innovation is 
happening today at the forefront of supercomputing.

   Fundamental technology enhancements are in the areas of 
        multi-core architectures with latency hiding techniques.

   Very fast interconnects that have the ability to make remote 
        memory access seem like local memory access

    Question 2. In your testimony you point out that ``while progress 
in supercomputing hardware has been astounding, both applications 
software and systems software remain a real challenge.'' Why do you 
believe this is so?
    Answer. Boosting system performance over the years--especially 
relative to price--has been comparatively easy, given Moore's law which 
posited a doubling of performance every 18 months. But the scope of 
these systems requires greater levels of coordination--to, among other 
things, orchestrate the tens or hundreds of thousands of processors 
that must work together in the world's most powerful computers. 
Further, the evolution of hybrid supercomputer architectures (The Los 
Alamos ``Roadrunner'' supercomputer, for example, will use both Cell 
and AMD Opteron processors) requires new software to take complex 
problems and divide their mathematical components for routing to 
different chip architectures simultaneously. On the application side, 
software development is required to extend the capabilities of high-
performance computing into commercial markets, as well as deeper into 
the academic and scientific arenas. The objectives of supercomputing 
hardware innovation are relatively uniform over time: more performance 
in less space requiring fewer/less resources. In the case of software, 
each new application requires new thinking, methods and skills.

    Question 3. Please explain how PC-based technologies have impacted 
the way that IBM develops its high-performance computers.
    Answer. Some of the world's most powerful computers are built using 
processors that are common in today's personal computers and video game 
systems. The emergence of parallel computing has allowed smaller, 
cooler, less powerful chips--such as those in PCs and game systems--to 
be grouped in large pools to split complex problems into smaller 
pieces.
                                 ______
                                 
    Response to Written Questions Submitted by Hon. John Ensign to 
                            Christopher Jehn

    Question 1. In your testimony, you state that, ``the supercomputing 
market is not large enough to justify significant investment in unique 
processor designs'' and that as a result, ``to advance supercomputing, 
industry has relied on leveraging innovation from the personal computer 
and server markets.'' Why is this development a bad thing?
    Answer. Industry's reliance on leveraging innovation from the 
personal computer and server markets has led to the production 
primarily of large collections of commodity processors linked together 
with commodity interconnects. This architecture dominates 
supercomputing. Unfortunately, these processor clusters are notoriously 
difficult to program. As a result, fewer programmers are capable of 
programming them, and those that do need much more time to program. The 
total cost of ownership goes up.
    Even more problematic is the divergence problem. See the following 
chart \1\ from a 2004 Federal interagency report on supercomputing.



    In looking at the chart, you can see that the theoretical peak 
performance, the highest performance achievable by a system performing 
at 100 percent efficiency, has diverged from the sustained system 
performance (SSP), which is what is actually usable by the 
applications. As you can see, the gap between the peak performance and 
the SSP has grown dramatically, meaning most of the system capability 
is wasted. As a result, supercomputers are far less efficient today 
than they were in 1996. And, the problem is getting worse. This 
development is at the core of the current crisis in supercomputing.

    Question 2. In your testimony, you state that there is currently a 
crisis in supercomputing. Several other witnesses seem to suggest that 
the situation is not quite that dire. Please elaborate on why you think 
that the situation is so problematic and what you believe needs to be 
done?
    Answer. Progress in advancing high-end computing technology has 
slowed considerably since 1996, when Federal Government R&D funding for 
supercomputing began to decrease. Supercomputers are now exceedingly 
complex and extraordinarily difficult to use and administer. 
Computational scientists now spend enormous amounts of time, effort and 
cost modifying software algorithms to run efficiently across large 
numbers of microprocessors with relatively weak interconnects. 
Programming and administrative costs often exceed the costs of the 
actual machine. Future trends in supercomputing will only exacerbate 
this problem.
    The lack of advancement in supercomputing technology has wide-
ranging ramifications. It not only puts our Nation's leadership in 
supercomputing at risk, but it also creates significant technology gaps 
that threaten our lead in national security, science and engineering, 
and our economic competitiveness. The most recent report on HEC, which 
comes from the U.S. Defense Science Board and the U.K. Defence 
Scientific Advisory Council, had this to say, ``There is great concern 
that lack of investment is eroding U.S. leadership in [supercomputing]; 
as well as, negatively impacting our ability to meet defense mission 
requirements . . . technology that is developed in the context of high-
performance computing `flows down' to help advance mass market 
computers. Thus, if the United States does not aggressively pursue very 
high performance computer technology, then innovation in mass market 
computers will slow.'' \2\
    Parallel programming is a prime example. Much has been discussed 
about how today's software for personal computers and workstations are 
not ready to run efficiently on multicore processors, even though 
computer vendors will soon stop selling computers with single cores. 
Had we continued Federal investments in supercomputing over the last 10 
years, it is likely large numbers of programmers would be able to 
program in parallel efficiently.
    The lack of advancement in supercomputing comes at a bad time. 
While HEC has been vitally important to the Federal Government for many 
decades, the need for supercomputing is greater today than ever. 
Federal agencies tell us this everyday. Agencies need supercomputing to 
help maintain military superiority, enable scientific research, advance 
technological development, and enhance industrial competitiveness. Just 
as they have in the past, Federal agencies rely on supercomputing to 
pave the way for real progress well into the future.
    As such, countries with the best high-end computing capabilities 
will enjoy a significant advantage in scientific competitiveness. 
Meanwhile, a number of other countries are aggressively pursuing HEC 
programs. Both Japan and China have announced programs to build multi-
petaflops systems in the near future. While little is known publicly of 
the Chinese plans, the Japanese government's proposed ``Keisoku'' 
project would spend US$1billion to build a 10 petaflops supercomputer 
by 2011 and another US$300 million for a new national HPC center.\3\ 
This aggressive government initiative is reminiscent of the Japanese 
Earth Simulator project whose performance took this country by surprise 
just 4 years ago.
    Our recommendation is that Congress fully fund the Administration's 
proposed government investments in supercomputing. This includes 
funding supercomputing programs in the Department of Energy (DOE), the 
National Science Foundation (NSF), and within the Department of Defense 
(DOD). Proposals from DOE's Office of Science and the NSF to fund the 
deployment of petascale computers are steps in the right direction. 
DOD, most notably through DARPA's High Productivity Computing Systems 
program and the National Security Agency (NSA), is supporting research 
and development to help reinvigorate the advancement of supercomputing 
technology. For example, the goal of the HPCS program is to provide 
economically-viable next-generation petascale supercomputing systems 
for the government and industry user communities in the 2010 timeframe. 
HPCS will significantly contribute to DOD and industry superiority in 
areas such as operational weather and ocean forecasting, analysis of 
the dispersion of airborne contaminants, cryptanalysis, military 
platform analysis, stealth design, intelligence systems, virtual 
manufacturing, nanotechnology, and emerging biotechnology.
    While these are important steps, we recommend the Administration 
build on these recent initiatives and develop and fund a coordinated 
research and development program for supercomputing, as many recent 
government sponsored reports have strongly recommended.\4\ The U.S./
U.K. Task Force on Defense Critical Technologies' recommendation to 
make HPCS a recurring program with multiple overlapping waves, each 
lasting seven to 8 years, is a sensible example of what such a program 
would include.
    We also recommend the Administration take a stronger leadership 
role in persuading other Federal agencies to make use of supercomputing 
and computational science to carry out agency missions. Many agencies 
have realized only limited scientific progress, because they are 
reluctant to complement experiment-based science with computational 
science.
    Numbers speak for themselves. While the NSF plans to spend more 
than $200 million on HEC out of its proposed $4.7 billion annual 
research budget, the National Aeronautics and Space Administration 
(NASA) plans to spend only about $30 million out of a proposed $10.5 
billion science, aeronautics and exploration budget. The National 
Institutes of Health (NIH) funding rate is even worse. NIH plans to 
spend significantly less then $30 million on HEC out of its $25.1 
billion annual research and development budget. The science and 
engineering requirements for NIH and NASA are not any less, dollar for 
dollar, than NSF's. Excluding NSA and DARPA, DOD plans to spend only 
$170 million on HEC out of its proposed $73.2 billion research, 
development, test and evaluation budget. That amounts to spending less 
than three tenths of 1 percent on high-performance computing, even 
though DOD's science and engineering requirements are enormous.
    Other agencies requiring significant science and engineering, such 
as the Environmental Protection Agency, the Department of 
Transportation and the Department of Agriculture, use practically no 
HEC at all. These agencies would benefit from the increased use of HEC. 
So, my final recommendation would be for the Federal Government to 
identify gaps in computational science usage across all of the agencies 
and develop programs to close these gaps where appropriate.
    Had the government continued investing as it had prior to the late 
1990s, we would more than likely have such promising technologies such 
as superconducting multiprocessors, processor-in-memory (PIM), 
multithreading, streams, and holographic storage today. We would also 
have seen similar advances in software.

ENDNOTES
    \1\ Interagency High-End Computing Revitalization Task Force 
Report--``Federal Plan for High-End Computing.'' May 10, 2004. http://
www.nitrd.gov/pubs/2004_
hecrtf/20040702_hecrtf.pdf.
    \2\ Joint U.S. Defense Science Board/U.K. Defence Scientific 
Advisory Council Task Force report--``Defense Critical Technologies.'' 
March 2006.
    \3\ IDC report--``The Keisoku Project: Reestablishing Japan's 
Leadership in Supercomputing?'' June 2006.
    \4\ National Science Foundation report, ``Revolutionizing Science 
and Engineering Through Cyberinfrastructure: report of the National 
Science Foundation Blue-Ribbon Advisory Panel on Cyberinfrastructure.'' 
January 2003. http://www.nsf.gov/od/oci/reports/atkins.pdf.
    Classified JASON's Report examining the requirements for 
supercomputing which derive from DOE's classified weapons research. 
Fall 2003.
    Department of Defense IHEC Report--``High Performance Computing for 
the National Security Community.'' July 1, 2002. http://
www.hpcmo.hpc.mil/Htdocs/DOCUMENTS/04172003_hpc_report_unclass.pdf.
    Interagency High End Computing Revitalization Task Force Report--
``Federal Plan for High-End Computing.'' May 10, 2004. http://
www.nitrd.gov/pubs/2004_
hecrtf/20040702_hecrtf.pdf.
    National Research Council Report--``Getting Up to Speed: The Future 
of Supercomputing.'' November 2004. http://www7.nationalacademies.org/
cstb/project_
supercomputing.html.
    The President's Information Technology Advisory Committee (PITAC) 
report--``Computational Science: Ensuring America's Competitiveness.'' 
June 2005. http://www.nitrd.gov/pitac/reports/20050609_computational/
computational.pdf.
    Joint U.S. Defense Science Board/U.K. Defence Scientific Advisory 
Council Task Force report--``Defense Critical Technologies.'' March 
2006.
                                 ______
                                 
     Response to Written Question Submitted by Hon. John Ensign to 
                              Jack Waters

    Question. Please explain how the manner in which research is 
conducted today reinforces the need for high-performance networks.
    Answer. Scientific research occurs across multiple disciplines, 
i.e., high-energy physics, biology, astronomy, etc. to name but a few, 
require significant super computing resources to process the 
extraordinary amounts of data. The research instruments, whether they 
are supercolliders, radio telescopes, electron-microscopes or 
laboratories, represent billions of dollars of public capital 
investment and continued operational investments by the public-sector.
    In order to leverage both the significant investment in research 
instruments and laboratories as well as the computational resources, it 
is essential that high-speed, high-capacity, high-performance networks 
be available to connect the disparate entities.
    To a very large extent, the research funded by the U.S. Government 
through various agencies, whether DOE, DOD, NIH, NSF, NASA, et al., is 
accomplished with multi-collaborating entities. Some, due to their 
highly sensitive nature and uniqueness are done with fewer 
collaborators, but even these tend to have multiple entities in 
disparate locations engaged in the research project. Simply put, 
supercomputers need to transmit enormous amounts of data amongst one 
another, hence the need for super networks.
    When you move into a research realm which is not as sensitive or 
unique, be it biomedical program with national defense implications 
like studying and preparing for a potential bird pandemic, a high-
energy physics program or astronomical research program, the number of 
collaboration partners can grow substantially to include hundreds of 
individual disparate entities.
    Networks are a key to facilitating research in order to truly 
leverage the investment made into widely dispersed physically entities. 
High-performance networks allow the sharing of computation resources 
minimizing the need for continued investment in supercomputers and 
accelerating the pace of scientific discovery.
                                 ______
                                 
    Response to Written Questions Submitted by Hon. John Ensign to 
                            Joseph Lombardo

    Question 1. How can a university like UNLV maximize the benefits 
from high-performance computing for its students and academic programs?
    Answer. A comprehensive university such as UNLV can maximize 
benefits from HPC for its students and faculty through interactive 
applications of real-time problems across the academic disciplines. 
Complex problems in research are struggling with the ever increasing 
amount of data to be processed (such as molecular biology in cancer 
research, difficult problems in dealing with nuclear waste, and 
emerging opportunities in the nano sciences). These problems are best 
addressed via management and analysis of data through applications of 
high-end computing. Developing an understanding of these research tools 
early in undergraduate education and hands-on applications by graduate 
students and faculty in real problems should enhance the capabilities 
of human resources and lead to advances in technological resources in 
the field of HPC across the board.
    University research programs could advance the applications of HPC 
through collaborative research programs with Federal agencies whose 
missions would benefit by use of HPC. Examples are the National Cancer 
Institute, the U.S. Department of Transportation (materials sciences/
new airport construction/development of sophisticated security systems) 
and the Department of Energy. The Committee might consider introducing 
resolutions or legislation encouraging such collaborations with various 
incentives to develop such consortia.

    Question 2. How does UNLV plan to update its high-performance 
computing capabilities as the next generation of hardware and software 
become available? Especially given the rapid pace of technological 
innovation in the high-performance computing area, how does a leading 
university like UNLV stay ahead of the curve?
    Answer. UNLV has embarked upon a program of developing symbiotic 
relationships with public-sector and private-sector groups that have 
need of HPC technology and the human resources that support the 
hardware and software infrastructure. These resources are sustained and 
enhanced by grants, contracts and other sources of funding that derive 
to the National Supercomputing Center for Energy and the Environment. 
These relationships and attendant contracts allow UNLV to stay current 
with the research needs and the requisite technology. Should 
opportunities emerge that would require a quantum leap in technology, 
UNLV might draw upon technological resources from other research 
universities (e.g., Ohio State University, University of California at 
San Diego, University of Alaska, et al.) as available and appropriate. 
Absent ability to access these extant resources, UNLV would seek 
Federal/state/corporate funding for the large capital outlays to 
enhance the technological and human resources.

    Question 3. In your testimony, you mention that, ``Support for 
Federal funding of high-performance computing has ebbed and flowed as a 
result of perceived foreign competition.'' Do you think that increased 
Federal support for high-performance computing research and development 
over the past few years is well-suited to meet today's foreign 
competition? Do you believe that data indicating that the United States 
possesses and produces a majority of the world's fastest high-
performance computers is a sign that we are out-computing and out-
competing our foreign competitors?
    Answer. The U.S. position in high-end computing is under constant 
global challenge as technological advances go forward. HPC is a vital 
tool to preserve U.S. pre-eminence in science, technology, math and 
engineering, and as such should be considered as an integral part of 
comprehensive national security. With the cutting-edge of technology 
always moving, it is necessary to the national interest that the 
Federal Government work collaboratively with university researchers, 
the corporate community and other related R&D interests to develop new 
approaches to the hardware and software technology infrastructure and 
to increasingly encourage applications of these research tools to areas 
that have not yet brought HPC to bear on their Grand Challenge 
problems. As Dr. Stan Burt, Director of the Advanced Biomedical 
Computing Center of NCI, pointed out in his testimony before the 
Committee, the biological sciences have lagged far behind the physical 
sciences in applying these 21st century tools to their research 
protocols. With increasing concern over biological warfare, 
communicable diseases, food safety and the traditional challenges of 
the fight against cancer, the Committee should make every effort to 
encourage the use of HPC in the field of biology and consortia fields 
of chemistry/physics/cellular studies, et al. Only with applications 
and access to HPC by all elements of the research community can the 
U.S. maintain its global leadership in HPC.
                                 ______
                                 
     Response to Written Question Submitted by Hon. John Ensign to 
                            Michael Garrett

    Question. From the perspective of a consumer of high-performance 
computing tools, how do you think the market for high-performance 
computers will evolve over the next 5 to 10 years?
    Answer. You have asked how we see the high-performance computing 
market evolving over the next 5-10 years. As far as the total market 
for high-performance computing we believe that the market will 
increase. We believe this for several reasons:
    Companies are moving to replace physical simulation with 
computational simulation (model the issue in a computer rather than 
build a piece of hardware and test it). The numerical simulation is 
both faster and cheaper--the issue is the fidelity of the answer. New 
applications for applying high-performance computing are emerging 
rapidly--such as genetic engineering, material modeling/creation (at 
the molecular level), weather prediction, entertainment and 
astrophysics. And finally, as the cost of high-performance computing 
comes down it will be more affordable to a broader range of customers.
    As for Boeing in particular our usage of high-performance computing 
will increase. The drivers for this are: (1) the need to get to the 
marketplace more rapidly. To do this we must shorten the design 
process, and the only way to do that is with more numerical simulation 
and less physical simulation. (2) We have a ``closet'' full of problems 
that are too big for current computers (at least affordable ones). As 
computer capability increases we will tackle these more complex 
problems.
    Another view of how the market will evolve is from a hardware 
perspective. We believe that the emphasis will be on large, cluster 
parallel-processing machines. Why? Because they use inexpensive 
commodity chips. Increased throughput will be achieved through faster 
processors and by applying more processors to the problem. Utilizing 
more processors presents software challenges as well. How will the 
application software be written so that it can run across 10, or even 
100 thousand processors? How will the operating system efficiently 
manage the data communications among these processors so as to not slow 
the process? And how will the operating system gracefully handle 
failures such that the entire process doesn't fail because a single 
element of the job does?
    The investment that the country makes in high-performance computing 
in support of nuclear stockpile verification, reconnaissance and 
intelligence gathering and processing, and for national defense in 
general will pay off in years to come as that hardware and software 
becomes available in the commercial marketplace. It is a model that has 
been in use for more than 30 years and as yet we do not see a reason to 
change it.
                                 ______
                                 
    Response to Written Questions Submitted by Hon. John Ensign to 
                         Stanley K. Burt, Ph.D.

    Question 1. In an era when scientific research is increasingly 
becoming inter-disciplinary, please discuss how the Advanced Biomedical 
Computing Center has adapted computational methods developed in other 
areas of science for use in biology research.
    Answer. The Advanced Biomedical Computing Center (ABCC) has, from 
practically its inception, had the good fortune to be composed of a 
staff that includes training in systems administration, network 
administration, bioinformatics, and physical sciences. As biology has 
become more dependent on computers and computational sciences, the ABCC 
has increased and diversified its staff in different scientific fields 
in order to utilize different scientific methodologies to solve 
biological problems. The diversity of the staff has allowed the 
creation of specialized tools, new algorithms, custom interfaces, and 
mathematical techniques. The ABCC is what I consider to be the model 
for a modern computing center for biology with the integration of 
different scientific disciplines combined with a strong computational 
infrastructure. The ABCC staff is knowledgeable about computer hardware 
and many simulation software packages and is able to adapt many 
techniques and computer architectures to specific problems. The ABCC 
also benefits from being in a diversified scientific environment that 
presents many kinds of biological problems demanding different 
approaches and solutions. The ABCC has benefited from the willingness 
of software and hardware vendors, other Federal agencies, and 
universities to work in a collaborative manner to find solutions for 
cancer.

    Question 2. What do you see as the greatest potential future 
benefits from using high-performance computing to assist in medical 
treatment of diseases like cancer or AIDS?
    Answer. I believe that the main role of HPC will be to tackle the 
inherent complexity of cellular processes and speed the time to 
solution for complex diseases and potential biological terrorism. As 
the increase in biological data accelerates at an incredible fast pace 
due to high-throughput screening, more genomes being mapped, and higher 
array chip densities, HPC will be needed to analyze the data, integrate 
the data, and model the complex processes involved in cancer and AIDS. 
No single experiment will be able to uncover these complexities. As 
biology moves toward a systems biology approach, which is necessary to 
use in order to understand the inter-relationships of complex cellular 
components in diseases, HPC will be necessary to address multiple 
levels of complexity at once, discover new data that will lead to more 
focused data-driven testable hypothesis, allow a better coupling of 
experiment and theory, and lead to more specific treatments. HPC is the 
best solution to provide a qualitative leap in understanding and 
solutions.

                                  
