Members:
Bill Joy, Co-Chair
Ken Kennedy, Co-Chair
Eric A. Benhamou
Vinton Cerf
Ching-chih Chen
David Cooper
Steven D. Dorfman
David Dorman
Robert Ewald
David J. Farber
Sherrilynne S. Fuller
Hector Garcia-Molina
Susan L. Graham
James N. Gray
W. Daniel Hillis
John P. Miller
David C. Nagel
Raj Reddy
Edward H. Shortliffe
Larry Smarr
Joe F. Thompson
Leslie Vadasz
Andrew J. Viterbi
Steven J. Wallach
Irving Wladawsky-Berger |
Honorable F. James Sensenbrenner, Jr.
Chair, House Science Committee
House of Representatives
2332 Rayburn House Office Building
Washington, DC 20515-4909
Dear Chairman:
Thank you for your letter of April 12 and your interest in the activities
of the President's Information Technology Advisory Committee (PITAC).
For the past year, the Committee has been reviewing the High-End Computing
and Computation (HECC) component of the Computing, Information, and
Communications (CIC) Research and Development (R&D) program in the
context of the entire Federal R&D investment in information technology.
Although the PITAC is concerned that changes in the structure of both
corporate and government support are putting US leadership in high-end
computing at risk, we see this as only one aspect of a larger problem--the
erosion of support for long-term fundamental research in information
technology. In the past, this kind of research has provided the foundation
for dramatic growth in the computing and communications industries,
helped address problems of national importance, and contributed to
the education of an entire generation of information technology professionals.
If the nation is to continue the rapid pace of advances experienced
over the past decade, we must reverse the drift away from fundamental
research.
Before exploring this issue in more detail, let us respond to the
specific questions raised in your letter:
(i) [Are] the overall level of resources for the HECC
component [of the CIC R&D program] adequate
and are they allocated appropriately?
Broadly speaking, there are two components to the HECC program, funding
for fundamental and applied research in high-end computing systems
and software and infrastructure to support high-end computation by
the national community of scientists and engineers. In our view, neither
of these components is receiving adequate resources.
One of the original goals of the Federal High Performance Computing
and Communications (HPCC) Initiative, the predecessor to CIC R&D,
was to extend U.S. technological leadership in high performance computing
and communications. Over time, many participants in that program came
to believe that the way to maintain leadership was to effect a paradigm
shift from vector supercomputers to scalable parallel computers, sometimes
called massively parallel processors (MPPs). Since scalable parallel
processing could build on commodity technologies, it could ride the
technology curve to achieve enormous processing power. Furthermore,
if applications were written to efficiently exploit parallelism, performance
would smoothly scale from desktop to supercomputer.
For this goal to be realized, however, there would have to be major
shifts in the way these machines were programmed to make effective
use of the available parallelism. Furthermore, there would need to
be innovations in hardware and software to provide adequate performance
of data movement between memory and the processors, and finally there
would need to be a new generation of programming tools to make it
easier to reprogram applications for scalable machines. To address
these needs, the program included a sizeable component of fundamental
and applied research support. Unfortunately, the level of research
support fell short of what was needed because (1) the problem proved
to be much harder than originally thought, and (2) as agency budgets
became tight, funds were shifted from long-term research to short-term
development efforts so some applications would be available on time.
The result is what might be called a "usability gap" for these machines.
Although there is every reason to believe that this gap can be bridged,
increased investments in research and development will be needed to
do so.
The usability gap also affects the availability of high end computing
cycles to the civilian science community. As a result, high end resources
available to NSF center users have not kept pace with their computational
needs, as evidenced both by the inability to satisfy requests, and
by the lower investment relative to the National laboratories and
university centers in other countries. This deficiency threatens the
nation's leadership in computational science and engineering.
Both of these areas need to be addressed if we are to continue the
rapid progress in the solution of important societal and scientific
problems through high-end computation. The Committee recommends that
the Federal government recommit itself to the support of high-end
computing through research on new computing architectures, technologies,
and software that can overcome the limitations of current supercomputer
designs. In addition, the Federal government should increase funding
for the acquisition of the powerful high-end computing systems needed
by the nation's science and engineering community, both within and
outside the government.
(ii) [Is] the HECC component coordinated effectively
with other high performance computing
activities of Federal mission agencies?
To be truly effective, coordination must take place both within the
HECC program and between HECC and other federal programs involving
high-end computing, such as DOE's ASCI program.
Within HECC, coordinating HECC-designated programs works well. The
coordination process, however, has only a limited ability to manage
the programs in ways that would lead to greater government-wide efficiencies
or to consider national benefits that transcend the specific needs
of contributing agencies. Furthermore, since a substantial fraction
of the HECC budget lies in mission agencies, resources tend to be
refocused on mission priorities over time, contributing to the drift
toward short-term applied research. To remedy this weakness, our report
recommends that an agency committed to fundamental research be designated
as a lead for coordinating information technology research including
HECC, with the mission to ensure that federal IT research is well
directed and with the flexibility and authority to carry out that
mission.
With regard to programs outside of HECC, we are particularly concerned
about the lack of coordination between HECC and the large computer
investments (both hardware and software) in the Departments of Energy
(ASCI) and Defense (HPC Modernization). To address this problem, these
and other similar agency programs current or under development,
such as DOE's Strategic Simulation Program -- should be coordinated
with HECC by the same mechanisms currently used within HECC.
(iii) [Will] the overall Federal R&D investment in high
performance computing
ensure U.S. leadership
in computing technology?
To maintain our leadership in high performance computing technology,
the United States must effect a transition from technologies that
have application exclusively at the high-end to strategies that are
more compatible with the commercial market. With the end of the cold
war, the market for the most powerful supercomputers has shrunk to
the point that it is no longer viable for a company to support itself
exclusively in that market. In the United States, the solution has
been to build on commodity technologies by using scalable parallel
computation. However, there have been significant problems with the
transition to these machines. The usability gap, alluded to earlier,
has slowed the translation of many applications to these systems.
In addition, there is evidence that some applications will not perform
well on these systems until bottlenecks in the memory hierarchy are
overcome by improved hardware and software. The committee believes
that it is imperative that we develop new, commercially viable computing
technologies to help solve the nation's critical problems. However,
we cannot move forward without overcoming the impediments to usability
of these technologies. Thus, we must recommit ourselves to a program
of increased research and development on high performance computing
architecture and software along lines similar to that proposed in
the recent HECC draft strategy. Such an increase will continue to
pay benefits to our nation by making it possible to build computers
of the highest performance largely from inexpensive but powerful commodity
technologies.
Over the past few years, there has been real erosion of Federal research
and development investments in information technology--HECC is only
one example. Left uncorrected, this erosion threatens the nation's
leadership in computing and communications technology and the economic
boom it has fostered. The Brooks-Sutherland report, "Evolving the
High Performance Computing and Communications Initiative to Support
the Nation's Information Infrastructure" (NRC 1995), documented a
clear relationship between investment in fundamental research in computer
and computational science and commercial impact a decade later. Foundations
for the current boom were laid a decade or more ago by the investments
from DARPA, NSF, and other government and industrial entities. The
erosion of government research funding, coupled with reductions in
long-term funding by industry and a continuing drain of university
faculty to industry, threaten to interrupt the stream of good ideas
that have fueled the information economy and led to the solution of
many crucial national problems in the 1990's.
To address this problem, the Committee recommends that funding for
fundamental research in information technology increase dramatically
over the next several years. A portion of that increase should help
reinvigorate research in high-end computing, including architecture,
software, and applications. In addition, the nation's non-defense
high-end computational infrastructure should be brought back up to
levels competitive with defense computing and with scientific computing
facilities in other nations, notably Japan and Europe. However, the
investment in computational infrastructure should be viewed as an
investment in instrumentation for science and engineering as well
as an investment in computing. In particular, it must be carefully
balanced against the goal of increasing fundamental research support
in all areas of information technology.
Thank you once again for your interest in our activities. The comments
in this letter are excerpted and adapted from our Interim Report to
the President, a copy of which is attached. Please do not hesitate
to contact either of us or any member of the Committee if you wish
further information or clarification.
|
