SUPERCOMPUTERS

• Document Type: CREST
• Collection: General CIA Records
• Document Number (FOIA) /ESDN (CREST): CIA-RDP85M00158R001000060021-4
• Release Decision: RIPPUB
• Original Classification: S
• Document Page Count: 51
• Document Creation Date: December 22, 2016
• Document Release Date: November 12, 2010
• Sequence Number: 21
• Publication Date: March 22, 1983
• Content Type: MEMO

Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Iq
Next 1 Page(s) In Document Denied
STAT
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
SECRET
DCI/ICS 83-4298-
22 March 1982L -
MEMORANDUM FOR: Director of Central Intelligence
Deputy Director of Central Intelligence
VIA: Director, Intelligence Community Staff
Director, Planning and Policy Staff
SUBJECT: Supercomputers
1. Action Requested: DCI/DDCI approval to brief IR&DC on 24 March on
current finings an to propose preliminary recommendations by the
Intelligence Community to the Chairman, IR&DC, and the OSTP.
2. Background: The attached briefing provides. a brief
status report on Intelligence Community plans tor supercomputers, views on USG
support to the private sector, and a set of informally coordinated
recommendations generated by this staff. The tone of the recnmmPndatinnc is
forceful. based largely
combine with our specific national security concerns. I
e ieve such a firm thrust is warranted.
The responses from the other members to the Keyworth Committee arrived
late last afternoon and are attached without comment. Significant to note is
the singular position taken by Bob Cooper, DARPA. I cannot support his views
and in fact I cannot even explain them.
Attachments:
(1)
(2) Responses from Keyworth
Committee Members
25X1
2bx1
225X1
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Approved by DDCI via telecon
with D/ICS 23 March 1983
Director of Central Intelligence ante
Director of Central Intelligence Date
SECRET
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
SUBJECT: Supercomputers
DISTRIBUTION: DCI/ICS 83-4298
Orig. - DCI (Return to PPS) w/atts.
1 - DDCI Watts
1 - Executive Registry w/atts
1 - D/ICS w/o atts
1 - ICS/PPS Subject w/atts
1 - ICS/PPS Chrono w/o atts
1 - ICS Registry w/atts
Attachments:
Series B, Copy 1,
Series B, Copy 2~
Series B, Copy 3,
Series B, Copy 4~
Series B, Copy 5,
DCI/ICS/PPS
Copy 1--DCI (Return to PPS)
Copy 1--DDCI
Copy 1--Executive Registry
Copy 1--ICS/PPS Subject
Copy 1--ICS Registry
22 March 1983 "supercomps" #24
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
UNLLA 11 1 W 11
~/ v7 0
ISSUE: How should the Intelligence Community respond to the challenge posed
worldwide supremacy in supercomputers?
UNCLASS I I-- D
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12: CIA-RDP85M00158RO01000060021-4
"SUPERCOMPUTERS"
DEFINITION: Fastest and most powerful scientific computing systems available.
Speed and capacity or special characteristics significantly greater than most machines for
commercial use.
CURRENT & POTENTIAL USES: Modeling or simulation of scientific or engineering problems in:
--
Intelligence: Cryptanalysis, COMSEC, signal processing,.economic modeling.
--
Defense:
Nuclear weapons design, ASW, weapon system command and control.
--
Science:
Weather prediction, aerodynamics, physics.
--
Commerce:
Oil exploration, microchip and computer design, aerospace'vehicle design.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 CIA-RDP85M00158RO01000060021-4
OUTLINE
INTRODUCTION
US SUPERCOMPUTER INDUSTRY
JAPANESE SUPERCOMPUTER INDUSTRY
STATUS OF US GOVERNMENT ACTIVITIES
ISSUES
PRELIMINARY FINDINGS
RECOMMENDATIONS
UNCLASSIFIED
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 CIA-RDP85M00158RO01000060021-4
F
IMPORTANCE OF US SUPERCOMPUTER LEADERSHIP
National Security users need an assured domestic source of supply.
,Supports maintenance of US technological leadership in key national security areas such as
cryptanalysis, COMSEC, and nuclear weapons design.
-- US Government needs served first.
-- Export control.
Supports US leadership in basic and applied science and in key high technology industries.
Provides prestige.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
INTELLIGENCE COMMUNITY QUESTIONS
What role should the Community play in USG assessments of the US supercomputer industry?
Should the DCI take immediate programmatic action in this area? How should these activities relate to
activities such as VHSIC, Artificial Intelligence?
SECRET
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12: CIA-RDP85M00158RO01000060021-4
US SUPERCOMPUTER INDUSTRY
Cray Research
Cray 1
Cray X-MP
Cray 2
Control Data
CDC 7600
Cyber 205
UNCLASS1F IED
Max Speed
80-160M FLOP
(3-6X Cray 1)
Early 70s 20-40M FLOP
-Late-70s- 800M FLOP r%/a?%A
2Q &
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 CIA-RDP85M00158RO01000060021-4
PARTIAL INVENTORY AND ANNOUNCED ORDERS OF CLASS VI MACHINES
(SUPERCOMPUTERS)
No. of Sites
Type**
Purpose No. Cray 1
No. Cyber 205
US
11
G
Weapons Research
11
--
5
G(2), I(3)
Fusion, Reactor Research,
3
2
3
G
Nuclear Power, Plant Design
Military, Intelligence
3
8
G
''Atmospheric Science, Aerodynamics,
4
4
6*
G(2), U(4)
Fluid Dynamics, Jet Engine Simulation
Research
3
2
5
I
3
2
4
I
Graphics, Timesharing
3
1
US Sub-Total
42
30
11
Germany
6
G
Research, Aerospace Research
3
3
France
4
G, U, I
Research, Nuclear Energy, Timesharing
4
--
England
7
G, U, I
Weather, Research, Nuclear, Petroleum
5
2
Japan
2
1
Research
2
--
TOTAL
61
44
16
*
**
Includes 1 Denelcor HEP.
G-Government, I-Industry, U-University.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
UNCLASSI9,'0
4
Trend in execution bandwidth
10' -4
C
0
V
a
X
W
0
J940 1950 1960 1970 1980
COMPUTING
1990
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
US SUPERCOMPUTER INDUSTRY -- PROJECTIONS
Market
- About $170 million per year.
- Projected Growth: 15% per year.
Speed (NSA Estimate)
- 25 X Cray 1 by Late 1980s.
- 200 X Cray 1 by Early 1990s.
JLV ET
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
JAPANESE SUPERCOMPUTER INDUSTRY
JAPANESE GOAL: Japanese Government (MITI) and computer industry want to be leaders in supercomputing.
Current Japanese Supercomputer Systems
High-Speed Computer Project
Fifth Generation Computer Project,
Competitive Factors
Summary
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Next 2 Page(s) In Document Denied
Iq
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12: CIA-RDP85M00158RO01000060021-4
HIGH SPEED LOGIC DEVICES
COMPUTING
10j "W
100,sW I mw
POWER DISSIPATION
UNCLASSo ftD
IOmW 100mW
h7F 1-4
TrA n5I5~Or
0
I Lm AWn e
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Next 2 Page(s) In Document Denied
Iq
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 CIA-RDP85M00158RO01000060021-4
OSTP AD HOC COMMITEE AND RELATED COMMUNITY ACTIONS
OSTP Ad Hoc Committee on Supercomputers formed via Keyworth memo of 13 December 1982.
24, January 1983 meeting of Ad Hoc Committee.
--I No consensus on approach (no strong commitment for USG involvement).
Questions for IC agencies (24 January).
-- Responses received 11 February 1983.
Second meeting of Ad Hoc Committee scheduled for 30 March 1983.
-- Review all inputs.
-- Develop consensus position on issues.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
ATTENDEES
(OSTP AD HOC COMMITTEE ON SUPERCOMPUTERS)
Chairman
I . 4
N. Q. Pewitt, OSTP
National Security
ICS
Robert Cooper, DARPA
Robert Kahn, DARPA
Alvin Trivelpiece, DoE
OSTP and NSF
Frank Eden, NSF
Andrew Pettifore, OSTP
Civil and Scientific
Raymond Colladay, NASA
Jack Kenebrock, NASA
Joseph Smagorinsky, NOAH
James Walsh, NOAA (Princeton)
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
STAT
 Sanitized Copy Approved for Release 2010/11/12 CIA-RDP85M001 58RO01 000060021-4
QUESTIONS FOR IC AGENCIES -- 24 JANUARY 1983
A. Do you currently own and operate or time share a supercomputer? How many or how much usage?
B. Do you have firm plans to acquire the next generation supercomputer? What specs? When? How many?
Approximate dollar funds per supercomputer?
C. Is there a difference in your requirement between a"big number cruncher" and an even bigger and more
complex Al-based machine?
D. Have you identified a US source or sources?
E. What impact would a successful Japanese fifth generation and supercomputer with the approximate
performance characteristics have on your agency?
F. Is there a role for the federal government in stimulating, partially funding, or actually developing
the fifth generation supercomputer? One body of opinion asserts the US Government should only indicate
the range of performance characteristic desired, the best estimate on quantity required and probable
time frame for delivery with a tolerable price range.
G. Do you have an opinion on the probable success or range of performance the Japanese are likely to
achieve? Basis for opinion.
ONCLASSED
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4 _
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
AGENCY VIEWS ON FEDERAL GOVERNMENT ROLE
DIA CIA
F. Federal government Provide incentives Support role for
rove? for consolidation federal gov't.in
& expansion of encouraging
' industry fifth domestic develop-
generation efforts ment
Yes.
- Provide funds to
support R&D.
- Stimulate gov't
industry/univ.
consensus on policy
& technology direc-
tions.
- Cryptanalytic input
to design effort.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Potential adverse effects on US computer industry competitiveness from:
-- Loss of technology spin-off.
-- Loss of leadership image.
Loss of industrial competitiveness from potential delays (or denial) to US scientific research
establishment and industry.
UI`~CL, ) I ILA
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
)L tL I
PROPOSED
Supercomputer Recommendations
1. The DCI support a role for the federal government in stimulating, partially funding and guiding the US
priOate sector in developing the fifth generation supercomputer and the DCI provide positive guidance
to tie Intelligence Community to begin studies for plans and programs to anticipate that capability for
each Program Manager as appropriate.
The DCI is encouraged to channel that support to the USDR&E who has established multiple programs in
device technology (VHSIC), systems and software (DARPA/NSA) and artificial intelligence throughout the
DoD and Intelligence Community. The USDR&E will, with appropriate Congressional approval and
legislation, establish a National Clearing House to collect and disseminate the results of USG-
SECRET
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
EXECUTIVE OFFICE OF THE PRESIDENT
OFFICE OF SCIENCE AND TECHNOLOGY POLICY
WASHINGTON, D.C. 20500
January 25, 1983
MEMORANDUM FOR MEMBERS FCCSET SUPERCOMPUTER PANEL
FROM: DOUG PEWITT
SUBJECT: Follow-up to January 24, 1983 Panel Meeting
The following are the three questions we agreed to consider and
to provide our individual agency views on for circulation by
the week of January 31:
i) What the Federal Government should do, if anything, to
meet the declared intent of the Japanese initiatives
to attain world leadership in the supercomputer field,
and to meet our own needs for very high speed,
general purpose supercomputers.
ii) How to insure access to existing supercomputers for
users whose present research computing needs cannot be
accommodated on less than Class VI machines.
iii) How to coordinate Federal R&D activities in artificial
intelligence, Fifth Generation computers, and informa-
tion processing related technologies generally.
We agreed that the last two questions might be appropriately
left to technical working groups, but we need to agree on
whether they are necessary, and if so, on our charges to them.
At our next meeting we will discuss proposed solutions to the
first question. Hopefully, we can agree on what the Adminis-
tration's response to this challenge will be, even if no common
strategy is advisable for all agencies.
Attached is a listing of attendees with addresses.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
COLLADAY, Raymond
Deputy Associate Administrator
National Aeronautics and Space Administration
Washington, D.C. 20546
COOPER, Robert
Director
Defense Advanced Research Projects Agency
1400 Wilson Boulevard
Arlington, Virginia 22209
STAT
Director
Office of Planning
Intelligence Community Staff
Washington, D.C. 20505
EDEN, Frank
Senior Science Associate
National Science Foundation
1800 G Street, N.W.
Washington, D.C. 20550
KAHN, Robert
Director
Information Processing Techniques Office
Defense Advanced Research Projects Agency
1400 Wilson Boulevard
Arlington, Virginia 22209
KENEBROCK, Jack
'Associate Administrator
for Aeronautics and Space Technology
Code R
National Aeronautics and Space Administration
Washington, D.C. 20546
PEWITT, N. D.
Assistant Director
Office of Science and Technology Policy
The White House
Washington, D.C. 20500
SMAGORINSKY, Joseph
Geophysical Fluid Dynamics Laboratory
National Oceanic and Atmospheric Administration
Post Office Box 308
Princeton, New Jersey 08540
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
TRIVELPIECE, Alvin
Director of Energy Research
Room 7B-058
Department of Energy
Forrestal Building
Washington, D.C. 20585
WELSH, James
Supervisory Computer Systems Analyst
Geophysical Fluid Dynamics Laboratory
Princeton University
Post Office Box 308
Princeton, New Jersey 08540
PETTIFORE, Andrew
Policy Analyst
Office of Science and Technology Policy
The White House
Washington, D.C. 20500
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
EXECUTIVE OFFICE OF THE PRESIDENT
OFFICE OF SCIENCE AND TECHNOLOGY POLICY
WASHINGTON, D.C. 20500
March 18, 1983
MEMORANDUM FOR DISTRIBUTION
FROM: DOUG PEWITT
SUBJECT: FCCSET Supercomputer Panel Feedback
Attached are copies of the responses you have each provided
so far, as agreed to at out last meeting.
It seems clear that the government has a role both as a
customer for the latest "off the shelf" systems, and as a
friendly host for supercomputer prototype evaluations. In
addition, the government has a major role in supporting, at a
minimum, supercomputer relevant, non proprietary research
and the associated training of future computer scientists.
Our next meeting has been scheduled for 3:00 p.m., March 30,
in Room 5026. At that meeting, I would like to discuss how
best we should implement these or. other appropriate roles.
This would enable us to reach a conclusion, on a plan of
action at least, to the first of the three questions we posed
at the last meeting. If there is time we will discuss how
best we should approach the subsequent questions of user
access and research coordination.
R. Cooper, DOD
ICS V
F. Eden, NSF
J. Kerrebrock, NASA
J. Smagorinsky, NOAA
A. Trivelpiece, DOE
STAT
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Department of Energy
Washington, D.C. 20585
1A,~~~ 3 1983
MEMORANDUM FOR: N. Douglas Pewitt
SUBJECT: Department of Energy Views on Supercomputers
Background
The Department of Energy's (DOE) laboratories have historically required the
use of the fastest computers available in order to carry out their respective
missions. Since the 1950's, a DOE laboratory has acquired the first or one
of the first manufactured units of nearly every new large scale computer.
These early models arrived almost devoid of usable software and consequently
the laboratories were required to provide the necessary systems and appli-
cation programs in order to fully utilize the new machines. Thus, DOE
laboratories have played an important role in causing the development of each
new generation of supercomputers.
Every generation of large scientific computers has had a significant impact
on the ability of the DOE laboratories to meet their mission goals. Each new
machine has improved the laboratories' abilities to more accurately model com-
plex physical phenomena thus improving their predictive capability and reducing
dependence on empirical scaling. This is particularly true in the design and
testing of nuclear weapons and more recently in controlled nuclear fusion
research. A recent inventory * of supercomputers (Cray I and Cyber 205)
indicates that 36% of the total number in the U.S. are located in DOE labora-
tories.
Until the late 1960's, a new computer generally provided an order of magni-
tude (10x) increase in capability. Since that time, however, the increase
has been more like fourfold (4x) with each new major main frame. At the same
time, the computational needs of the DOE laboratories have increased at a
faster rate. Consequently, supercomputer technology has fallen short of
meeting present desired capability. Today,DOE laboratories have major pro-
grams that have a strong need for a minimum of a 200x increase in capability
by the end of the decade. Current hardware development schedules by U.S.
manufacturers further indicate that far less than this improvement will be
achieved by 1989 unless a significant major increase in development effort
occurs in this country.
*Report of the Panel on Large Scale Computing and Engineering,
Peter D. Lax (Chairman), NSF, December 1982.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Although scientists and engineers in DOE's weapons and magnetic fusion pro-
grams have ready access to state-of-the-art (Class VI) supercomputers,
scientists in other DOE programs such as Basic Energy Sciences, High Energy
Physics, Nuclear Physics and others do not have such access. A similar
situation exists in research and development programs supported by other U.S.
government funding agencies. Recent studies * suggest that many of these
R&D programs would benefit substantially from the use of supercomputers. Lack
of access by universities causes a special problem since there is now very
little training of new scientists and engineers in the use of supercomputers.
In DOE, the Office of Energy Research is presently considering alternatives
for providing supercomputer access for its scientists. As an interim solution
800 hours/per year of Cray 1 time on the Magnetic Fusion Energy Network has
been made available to other scientists in the Department's Energy Research
programs in DOE laboratories and universities.
Both the application of supercomputers to present R&D programs and the
development of new generations of supercomputers are important to the future
leadership of the U.S. in defense technology; in research and development; and
in many commercial areas as well. Perhaps one of the most important issues
is the national security implication of buying supercomputers from a foreign
country if U.S. industry were to lose its leadership in this critical area.
For these reasons, we believe that the U.S. government should take appropriate
steps to encourage application of present supercomputers to a variety of
scientific and engineering problems and to stimulate U.S. computer vendors to
accelerate development of next generation machines.
Recommended Actions
1. Accelerated Development of New Supercomputers.
Should the U.S. government play a role in the development of supercomputers?
There are several facts about the supercomputer industry that are important
to this discussion. The present supercomputer market is relatively small at
about $lOOM per year. It appears that most of the large computer manufacturers
have decided not to enter this market segment presumably because of its small
size. Only two companies, CDC and Cray Research, have produced Class VI
machines.
Start-up and R&D costs could range from $50-100M for a new generation of
supercomputers. Cray Research is a relatively small company and Denelcor,
another vendor which could potentially become a force in the supercomputer
market is a very small company. The R&D capabilities of both Cray and
Denelcor are limited and it is questionable whether they can compete with
government subsidized vendors in other countries in the long term. In sum-
mary, the large computer companies seem to be reluctant to devote substantial
R&D funds to supercomputers and the smaller companies have very limited
resources for these developments.
*Report of the Panel on Large Scale Computing and Engineering,
Peter D. Lax (Chairman), NSF, December 1982.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Another important fact is that the U.S. government is a large consumer of
supercomputers having purchased or leased over 50% of all of the Class VI
computers currently in service in the United States. The government has
played a significant role in developing useful supercomputer systems in
the past to meet its own needs. For example, the DOE accepted the risks
by buying the first production Class VI machines and the DOE laboratories
did substantial software development as they have historically. Other
customers started purchasing Class VI machines only after they had been
demonstrated in DOE laboratories.
In view of the U.S. government's need for advanced supercomputers, the
limited development capabilities of some key vendors, and the historical
government role in development of supercomputers, several government actions
are recommended. If the government plays the proper role, it could stimulate
competition within the United States as well as ensuring the continuing
dominance of the United States in future supercomputer markets.
Both Cray Research and Control Data Corporation are developing advanced
machines. The first of the new Cray-2 machines is expected to be available
in early CY 1985. The CDC machine is expected sometime later. For the near
term, the government should continue to play the role of purchasing early
production models and its laboratories should continue to assist in the
development of software for those production models.
For the generation of machines beyond the Cray-2, the government should
provide incentives that will accelerate supercomputer development in order to
meet its own needs and to help ensure continued U.S.. leadership in this criti-
cal field. A goal of at least 200x present capability by 1989 appears
feasible, but attaining this without government incentives is unlikely. The
target machines would be commercially viable, with a selling price of $10-15M.
The Japanese have formally created a national program to achieve a super-
computer with 100x present capability by 1990.
The program proposed for the U.S. government is based upon a phased, competi-
tive procurement. The details of this procurement have not been worked out,
but it might have three phases as described below.
The first phase would solicit proposals to generate paper specifications from
a number of potential vendors. This phase would last one year and might
involve 5-6 vendors. The product of this phase would be a number of paper
designs of supercomputer systems, including architecture, technology, and
software. Evaluation of this phase would benefit from the experience gained
with experimental machines, discussed in IB.
During the second phase, the most promising designs (perhaps three) would be
supported for prototype hardware and software development. The second phase
would last two years and the government would receive the right to use the
prototype.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
During phase three, those prototypes which were judged capable of attaining
the performance goals would. undergo final development and initial production.
This phase, lasting two years, would also include experiments using the proto-
type machines by interested government laboratories in close cooperation with
the vendors. Government funding of phase three would be through a guaranteed
purchase of approximately two machines per vendor with the expectation that
there would be additional purchases by both government and industry. Through
this mechanism both the hardware and software may be modified to better suit
the end user's need.
At the end of the approximately five year development cycle, the new
machine(s) becomes fully commercial. If this process is successful, it is
envisioned to lead to a semi-continuous process that could be repeated for
the following generations of supercomputers.
B. Government Supported Research and Development
In addition to the procurement activities described, which will stimulate
industrial R&D, some specific government supported R&D is recommended. The
government should selectively provide early financial support to innovative
architectural and technological approaches taken by private industry. This
would help to ensure survival of such approaches until they can become
marketable as production solutions to known requirements. It is also needed
to encourage private industry to engage in high risk R&D ventures which may
accelerate the advancement of supercomputers.
One approach the government should consider is the immediate establishment
of an R&D program(s) to purchase experimental systems such as the Denelcor
Heterogeneous Element Processor (HEP) and the CDC Advanced Flexible Processor
(AFP). Several of these experimental systems should be purchased and installed
in various federal laboratories and made available for research purposes to
universities and industry as well as the laboratories. This would permit
research on the use of advanced architectures as they relate to new languages,
algorithms, software tools and applications systems.
Care should be taken to assure that a clear distinction is made between the
acquisition and use of such experimental systems versus the acquisition and
use of production computers to avoid misunderstandings concerning the expec-
tation of results.
Future increases of orders of magnitude in supercomputer speeds can be
achieved only by incorporating multiple processors and perhaps hundreds or
thousands of processors in new parallel architectures will be required. For
approximately the next five years, the appropriate role for DOE is to support
long term research in languages, algorithms, and software for parallel
architectures. Government agencies should provide support for: faculty,
staff, and graduate students; access to commercially available experimental
systems; and several university/industry/laboratory collaborations,to build
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
and experiment with new experimental hardware/software systems. Increased
support for some of these longer-range research activities will be required
if they are to impact the development of commercial machines in a timely
fashion.
Since the DOE is the largest user of supercomputers and has traditionally led
in advancing the state-of-the-art in numerical algorithms and software, the
DOE should remain the lead government agency in accelerating the development
of advanced supercomputers and in exploration of large scale scientific
computational techniques.
II. Providing access to supercomputers for a larger number of scientists
and engineers.
Providing access to present and future supercomputers for a larger number of
U.S. scientists and engineers is an important step that will provide several
benefits. Supercomputers are important new scientific tools which have the
potential to advance rapidly and substantially research and development in
many fields. Making supercomputers available to researchers in universities
has the additional benefit that students will become familiar with their use.
The training aspect is important because presently very few scientists and
engineers graduate with any experience in using this modern scientific tool.
Also, greater use of supercomputers will undoubtedly increase their appli-
cations and should help bring solutions to many problems where their use
may not now be anticipated. These increased applications would expand the
supercomputer market.
In DOE's magnetic fusion program, which does provide access to Class VI
computers for all its contractors, substantial benefits have been derived.
Other programs and other agencies should carefully consider the potential
benefits of greater access to existing supercomputers for researchers in
their R&D programs. This should be an individual agency decision. As
mentioned above, DOE's Office of Energy Research is currently planning to
provide access for more of its scientists in DOE laboratories and univer-
sities.
A.FCCSET working group should be formed to discuss potential interagency
cooperation and to share experiences and plans in this area.
III. Coordination of Federal R&D Activities in Artificial Intelligence,
Fifth Generation Computers and Information Processing.
There is currently research and development in advanced computer architecture,
hardware, software, etc., funded by several government agencies. It would
undoubtedly be beneficial to coordinate this R&D across the government.
An interagency committee of working level people should be established to
coordinate activities in advanced computer development.
Alvil{W. Trivelpi
Director, Office of
Energy Research
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Dr. N. D. Pewitt
Assitant Director, OSTP
The White House
Washington, DC 20500
U.S. DEPARTMENT OF COMMERCE
National Oceanic and Atmospheric Administration
ENVIRONMENTAL RESEARCH LABORATORIES
c/o GEOPHYSICAL FLUID DYNAMICS LABORATORY
Princeton University - P.O. Box 308
Princeton, New Jersey 08540
Phone: [phone], FTS and Commercial
February 9, 1983 R/E/GF
SUBJECT: NOAA Recy'mmendation for Supercomputer Policy
In my opinion the first of the three questions
enumerated in your January 25, 1983 memo can be separated
from the last two, which can be dealt with at a later time.
I feel that question (ii) is important, but I do not
presume to contribute much to question (iii).
My views on Japanese competition are briefly addressed
in section A, below, in the form of questions that it
raises on motivation and ultimate objectives. The
remaining sections discuss NOAA's needs and procurement
practices as they may be relevant to an evolving
Government-wide policy. They are proposed for
consideration to stimulate the best and most timely
response from the industry, whether the market place be
world-wide or confined to American vendors.
A. U.S. Industry vs. World Competition. There is
little question that the U.S. Government is the major user
of top-of-the-line computing systems and will continue to
be so. It will therefore need from the industry at-large
the maximum rate of improvement in power at the least
possible cost. To this end the U.S. Government should
assure a healthy competition in top-of-the-line computers.
Whether the U.S. Government seeks such a yield
exclusively from American industry depends on several
factors:
* Will it inhibit constructive competition in the form
of rapid innovation and minimum price?
* Will it insure self sufficiency (in a national
security sense) under a less friendly political situation?
* Is it useful to distinguish an open market for
components only (e.g., I.C. chips) from an open market for
integrated main frames?
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
Doug Pewitt; February 9, 1983; Page 2
* Should Government preference for American vendors be
in the form of protective tariffs on Japanese imports or
just in the form of a subsidy for American industry to
stimulate development?
* If the Government subsidizes American industry,
should it be through a direct subsidy (the way the Japanese
do it) or indirectly by a guarantee to acquire a minimum
number of early copies of each generation (the way AEC did
it in the 1950's, 60's and 70's).
B. The Scope of Meteorological Needs. The use of
supercomputers in meteorology historically goes back to
some of the first research studies on the computer built at
the Institute for Advanced Study (IAS) in Princeton in the
early 1950's. One of the predecessors of NOAA, the U.S.
Weather Bureau, jointly with the Air Force and Navy, soon
after acquired an IBM 701 for operational short range
numerical weather prediction to mark the beginning of an
unbroken sequence of top-of-the-line computer acquisitions.
By next year NOAA will have three CYBER 205's, dedicated to
meteorological and oceanographic research and to
operational weather prediction.
This demand by the geophysical fluid dynamics sciences
is further reflected in the USA by the National Center for
Atmospheric Research (Cray-1), which is NSF supported, by
partial use at the NASA Goddard Space Flight Center (CYBER
205) and by the DOD (CYBER 203, upgraded to a 205). A few
universities also serve academic meteorological research
with supercomputers, e.g., the Colorado State University
(CYBER 205). Abroad the U.K. Meteorological Office uses a
CYBER 205, and the Canadian Meteorological Service and the
European Center for Medium-Range Forecasts each has a
CRAY-1. The Japan Meteorological Agency now has at least
two top-of-the-line HITAC-M200H computers. One of these is
scheduled to be replaced in the near future by a more
advanced model. The Meteorological Department at Tokyo
University has access to the University's HITAC computer.
In short, the meteorological research and operations
community has become a major market for supercomputers both
domestically and world-wide. Most calculations to
integrate non-linear systems of three space-dimensional,
time-dependent systems of partial differential equations
are still severely resolution limited. Since just a
doubling of resolution in each of the three space
dimensions generally entails a 16-fold increase in the
amount of computation, it is likely that such uses will
continue to be at the forefront of demand for
supercomputers of the future. This, of course, is a direct
counterpart to the needs of "computer wind tunnel"
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
Doug Pewitt; February 9, 1983; Page 3
applications. The main difference is that the number of
systems needed in meteorology and oceanography may be an
order of magnitude greater.
C. Software. There is fundamental significance in the
use of the term "system", in that software capability can
influence throughput by as much as an order of magnitude.
The term "supercomputer" is not well focused. It seems to
draw more attention to raw hardware speed than to
programmability. It is therefore important to include the
significance of software in the operative description, for
almost all current supercomputer applications are, in fact,
programmed in Fortran and run on more-or-less conventional
operating software. The computer system may well be termed
"Fortran engine". Software specifications to bridge the
gap between hardware improvement and enhanced user needs
are an integral element of system upgrading, whether the
software is provided by the hardware vendor, by another
commercial vendor or by the user itself.
D. Industry Responsiveness: Off the Shelf vs.
Subsidized Development. NOAA has historically favored the
normal commercial development of supercomputers. The most
effective policy that can be followed by the U.S.
Government is to assure a stable marketplace for such
machines, although a stable, better-defined marketplace
will not guarantee continued U.S. strength. The principal
reason that the current supercomputer marketplace is not as
effective as possible is the great inefficiency of the
Federal procurement process and the balkanization of
requirements. Each procurement of a supercomputer is a
separate, isolated process that takes typically 4-5 years
from requirements to delivery. Because each Federal agency
must completely perform its own procurement, its
specifications tend to be particular and self-serving,
which thwarts standardization and cannot benefit from the
clout of potential multiple procurements.
NOAA would be well served by a Federal policy that
calls for annually competed supply contracts for
top-of-the-line computing systems. By eliminating the
RFP-Proposal cycle, agencies should be able to acquire such
systems in 2 years, instead of 4-5. A top-of-the-line
computing system is not expected to stand alone; a similar
procurement mechanism should be in place for a time-sharing
system that can serve as a front end.to such computing
engines.
A government-wide supply contract for top-of-the-line
computing systems would have the following advantages:
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
Doug Pewitt; February 9, 1983; Page 4
* By reducing the procurement cycle significantly, the
Government will benefit more rapidly from technological
advances and greater efficiency available in the
marketplace.
* Hardware and software standards can be more easily
enforced both on vendors and on the agencies.
* Vendors can anticipate a predictable, steady market.
* Vendors will have as a design target a standard
functional architecture that is based on Fortran.
E. Maximum Bang for Buck vs. Minimum Price. It is
virtually axiomatic that a top-of-the-line customer wants
the most powerful system available at the time. It is of
interest and significance to note that comparison of
today's supercomputer (say the CYBER 205, or CRAY-1) with
that of 30 years ago (the IBM 701) shows less than a factor
of three increase in inflation-adjusted dollars. For this,
the present generation is 10,000 times more powerful, has
1000 times more high-speed memory, has almost limitless
disk storage, and possesses an extremely sophisticated
software ensemble which is about half the cost of the
system.
It is proposed that supply contracts for
state-of-the-art computer systems be awarded as the result
of annual competitions. Agencies would then be free to
place orders subject only to the availability of funds.
For the agency the entire RFP-proposal-award cycle is
bypassed. Delivery should be possible within 12 months
after receipt of order.
The basic idea for evaluating competing systems is to
select the system that can demonstrate the most
productivity on a standard benchmark suite. The method of
aquisition for evaluation should be lease. One of the
goals of the streamlined procurement process is to
facilitate more frequent turnaround of systems, so that the
Government is always exploiting the latest technology and
efficiency as much as possible. Whereas the conventional
procurement fixes workload and competes on lowest price;
the strategy for computer systems should fix price and
compete on highest productivity. Fixing the price of
supercomputers facilitates long-term budgetary and
programmatic planning throughout the Government.
NOAA's recent procurement that resulted in a system
with two CYBER 205's was conducted on this basis of most
productivity for a specified annual budget. The only
problems we have encountered are associated with relaxing
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
Doug Pewitt; February 9, 1983; Page 5
standards and requirements to assure a basic level of
competition (two vendors) for the procurement. In any
government-wide competition any compromise of standards
must be strictly avoided. It is precisely through the
standards set for competition that the Government can most
effectively influence the future of supercomputers.
F. Shared Risk. A fundamental problem with
top-of-the-line computer systems is that the meeting of
design specifications is unknown until after delivery.
Such systems are often not available for actual testing at
the time that the contract is let. A competitive choice is
normally based on indirect preliminary inferences and
claims by the vendors. The Government must be protected
against irresponsible or inaccurate initial claims by not
accepting the complete risk for non-performance.
Contractual provision must be made for adjustments to be
levied upon the vendor for not meeting specification
claims. NOAA has successfully imposed such contractual
constraints on two procurements during the past decade at
GFDL: a Texas Instruments ASC/4X delivered in 1974 and a
CYBER 205 system delivered in 1982. The T.I. contract
provisions were actually enforced for non-performance and
the CYBER procurement is now similarly being tested.
cc: Dr. A. Calio, Deputy Administrator, NOAA
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
IVI'/'
National Aeronautics and
Space Administration
Washington, D.C.
20546
Reply to Attn of:
RT
Mr. Doug N. Pewitt
Assistant Director
Office of Science and Technology Policy
The White House
Washington, DC 20500
Subject: NASA (OAST) Position on Government's Role In
Support of General Purpose Supercomputers
This is our response to point one of your inquiry of January 25,
1983.
We believe the government should have three distinguishable roles
in furthering the development and utilization of supercomputers
in the U.S.: (1) encouraging private development of advanced
supercomputers by acting as a supportive first user, (2)
supporting research in computer science and technology, and (3)
assuring U.S. manufacturers free competitive access to inter-
national markets.
The first two roles are currently recognized by government
agencies, however increased emphasis is warranted in light of the
current technical opportunities and international competitive
situations. The third role should be reexamined, in the context
of overall U.S. trade policy.
Encouraging Development of Advanced Supercomputers
By acting as first users of prototype supercomputers, federal R&D
agencies can assure early evaluation of technical opportunities
and assist in the early recovery of private R&D investments.
The prototype hardware and software will enhance the user
agencies' capabilities in R&D and its use as early as possible
will stimulate beneficial interaction between the user community
and the developer. Enhanced capability to address many technical
problems of critical importance will result. Since development
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158R001000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
costs for prototype hardware and systems software are low
relative to those in many other high technology industries, rapid
recovery of a portion of development cost through early sales to
government R&D agencies should stimulate the rate at which new
computers can be introduced. To assure ready acceptance of new
products by commercial users, it is essential that the government
work with the manufacturers to mature prototype designs into
commercial systems.
The government should encourage applications of supercomputers by
supporting research leading to new algorithms and numerical
methods for the efficient solution of large engineering and
scientific problems. Such research, motivated by fluid
mechanics, structures and dynamics, chemistry, and physics, can
stimulate existing markets for supercomputers, and develop new
ones. Advances in computational fluid dynamics can lead to the
use of the supercomputer in the design and analysis process for
complete aircraft; advances in computational chemistry could lead
to the use of the supercomputer in the chemical industry to
develop synthesis processes and to select catalysts to control
reaction rates.
The government should increase its utilization of supercomputers
to meet unique government needs in areas such as weather
prediction, defense system analysis, nuclear energy, and space
imaging and data analysis. Such applications can provide a
market for increased numbers of supercomputers, while fostering
the aims of the user agencies.
Finally, the government should provide low cost supercomputer
time to critical industrial and academic users, who otherwise
would be unable to affort it. Focal points within the government
should not only provide access, but also provide supporting
expertise so that new users can efficiently utilize the computer
in the solution of their problems. Such use will not only
encourage additional applications, but also provide feedback to
the computer system developers of potential computer hardware and
software advances which will enhance capabilities.
Computer Science Research
The government must take a strong role in support of academic,
government, and industrial research in computer science and
technology. The thrust should be to accelerate the solution of
technical problems associated with components, system
architectures, algorithms and software for advanced super-
computers. Research should range from basic phenomena through
proof-of-concept prototypes.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
Remove Constraints to International Trade
Competition with foreign producers of supercomputers is adversely
affected by restricted access to foreign markets by U.S. super-
computer manufacturers. Industry contends that an expanded
market provides R&D finance essential to assure their ability to
compete with foreign manufacturers. There is a compromise to be
made between restricting the export of production supercomputers
to limit foreign nations' exploration of their capabilities and
improving the economic health of the domestic industry to help
preserve its leading position in technology; Foreign sales of
U.S. production supercomputers are sometimes restricted for
strategic reasons by the U.S. government, and the voids will soon
be filled by the development of supercomputers of foreign design
and manufacture. These developments are fostered by foreign
government investment with their computer industry. The
government should reassess its policy toward trade restrictions
to consider the value of maintaining a technological lead in
commercial supercomputers by assuring a larger foreign market
versus restricting sales to a smaller member of strategically
safe countries and allowing entry of foreign developed super-
computers into the market place.
Jack L. Kerrebrock
Associate Administrator for
Aeronautics & Space Technology
v Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
DEFENSE ADVANCED RESEARCH PROJECTS AGENCY
1400 WILSON BOULEVARD
ARLINGTON VIRGINIA 22209
JAN 31 1983
SUBJECT: FCCSET Supercomputer Panel Comments
The comments below reflect the views of the Defense Advanced Research Projects
Agency on the three questions addressed in your memorandum of 25 January. I
believe they also reflect, generally, the views of Defense but, in view of the
short suspense, we have not had time to coordinate them.
i) The government should continue to provide support to agencies such as
DOE and NASA which are heavy users of general purpose supercomputers
so they may buy them from industry at a rate which is sufficient to
meet their needs and to retain a U.S. capability in the field.
Except for specialized NSA requirements, Defense has not been a major
user of conventional supercomputer technology. Defense anticipates
requirements for a new kind of very fast symbolic computing
capability which relies heavily on artificial intelligence and for
which significant research is still needed. Industry does not
currently have the know-how to build machines of this type.
ii) The existing class VI machines may not have sufficient capacity to
support all the users whose present research cannot be accommodated
with less computing power. A survey should be conducted to determine
the major class VI requirements within government, industry and
university groups carrying out state-of-the-art scientific research.
It may be appropriate to authorize the purchase and operation of
additional machines over a suitable time frame for shared use by the
community. These machines could be made more widely available by
connection to an appropriate network subject to appropriate
administrative controls on access to the machines. The National
Laboratories at Livermore and Los Alamos may be the appropriate place
to house this capability.
iii) At the present time, there are few if any major development programs
in government or industry in artificial intelligence or 5th
Generation computers. If, as expected, R&D in this area should
increase significantly over the next ten years, existing coordination
groups such as the Interagency Committee on Extramural Mathematic
Programs (ICEMAP) would be candidates to include it. NSF and DoD
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
officials also meet regularly to discuss R&D programs in both
organizations and this would be another forum for increased
coordination. No further coordination action appears warranted
in this area at the present time.
Robert S. Cooper
Director
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
DEFENSE ADVANCED RESEARCH PROJECTS AGENCY
1400 WILSON BOULEVARD
ARLINGTON VIRGINIA 22209
FEB - t 1983
Dr. N. D. Pewitt
Assistant Director
Office of Science and Technology Policy
The White House
Washington, D.C. 20500
Dear Doug:
This is a follow-up to your request for comment on the three questions on
Supercomputers which you enunciated at the meeting last Monday. Although Bob
Kahn prepared an "ARPA" response (which you will receive under separate
cover), I thought I would take the opportunity to expand on these important
questions personally.
From a Defense (historical) perspective the market for Class VI and beyond
number crunchers is very limited. As I pointed out at Monday's meeting,
Defense has not taken advantage of such machines historically except in an
indirect way through the DOE weapon design community and through the intelli-
gence agencies. Instead the vast majority of Defense fixed base computer
acquisitions have been dfor less capable machines that have been used for date
base manipulation and C functions.
Although the scientific and engineering uses of Class VI machines would
benefit the Defense Research community, defense research managers have not
felt that they could justify the levels of funding ($l0's of millions) in
their budget advocacy process to buy supercomputers. Hence, they have made no
serious attempts to acquire them. What is more, the barriers thrown in the
way of computer acquisition by Federal Regulations, GSA and Congressman Brooks
have had a chilling effect on all scientific computer acquisition overall.
A factor in the disinterest shown specifically by the Defense labs in. Class VI
computing machines is the lack of qualified personnel in the labs to use such
machines. Government personnel rules including the ones regulating general
schedule employees and the senior executive service employees salaries, perks
and working conditions have driven many of the best persons in the computer
science and engineering fields out of government.. The labs no longer have
extensive capability to provide high quality scientific talent and computer
code for machines such as the CRAY-1 which is delivered with very rudimentary
software systems.
The net conclusion I draw from all. of this is that the Japanese could sieze
the entire market for supercomputers and it would affect the Defense Research
community not at all. Furthermore, in order to create an interest in super-
computers and a highly qualified user community in Defense Research, we would
have to solve some nearly intractable problems of regressive Government
personnel policy, restrictive federal regulation of computer acquisition and
unbridled turfsmanship by GSA and Congress. These are probably worthy prob-
lems to be attacked by the "Man of LaMancha."
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
My view on your second question is that we have not provided appropriate
computing resources across the board for the university scientific community
including general purpose supercomputers. Both NSF and Defense have plans
over a several year period of time to upgrade the computing machines generally
available for basic research in the computer community. However, there are no
general plans to acquire and make available Class VI and beyond machines..
They are just too expensive and hence do not relate well to our basic research
programs. Only a few specialized cases such as that at NCAR and at Ames
Research Center have actual plans related to G.P. Supercomputers being devel-
oped.
I feel that the Nation would benefit from policy level support for supplying
computer cycles to basic research institutions. It could help NSF and DARPA
get both Administration support (0MB, SecDef, etc.) and congressional support
for our programs. In addition, I believe a specific recommendation to make
the CRAY machines at LASL and L3 available over a network to qualified
researchers is in order. Both of the computer departments at these institu-
tions are favorable to such an arrangement. If a survey reveals that legiti-
mate demand for cycles is greater than supply, then DOE's budget should be
augmented to acquire more resources to meet the demand.
On the question of so-called fifth generation computers and artificial intel-
ligence research, it is probably too early to augment the existing coordina-
tion mechanisms on programs in this area. Little real program activity is
planned in this area prior to FY84 and even then the ideas upon which new
programs will be based will not be clear until FY85. It appears that the
right strategy here is to let plans develop under existing mechanisms for
about a year or so then take another look to see if something more is needed.
I hope these views are useful in your consideration of actions needed.
Sincerely,
Robert S. Cooper
Director
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
NATIONAL SCIENCE FOUNDATION
WASHINGTON, D.C. 20550
February 7, 1983
OFFICE OF THE
ASSISTANT DIRECTOR
FOR ASTRONOMICAL.
ATMOSPHERIC, EARTH.
AND OCEAN SCIENCES
MEMORANDUM
TO: D. N. Pewitt
Assistant Director, OSTP
FROM: H. Frank Eden ',1, '
NSF Member FCCSET Supercomputer Panel
SUBJECT: NSF Views on Supercomputer Topics
This memo outlines the NSF view of the three topics raised at the panel
meeting of January 24 and in your memo of January 25.
Background: NSF Involvement
NSF is involved in large scale computing in two principal ways:
- Research in computer science and engineering is supported through
the Divisions of Mathematical and Computer Sciences, Electrical,
Computer and Systems Engineering and Information Science and
Technology. Topics supported cover the entire spectrum of basic
research in computer science and computer engineering.
- Through the National Center for Atmospheric Research (NCAR), NSF
has made available to atmospheric and ocean scientists a computing
facility based on the most advanced machines available over the
past 20 years. Until 1983 the facility used a CDC 7600 and a
CRAY I. Recently NCAR acquired a second CRAY I as a replacement
for the CDC 7600 and the facility is converting to use both CRAY
machines. The facility services some 900 visitors and 75 remote
terminals across the Nation. Plans call for the acquisition of
a Class VII machine to replace one of the CRAY I machines over
the next few years.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85MOOl 58RO01 000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
- In addition, researchers in disciplinary areas requiring large
scale computation can include computing costs in their proposals
for NSF support as appropriate. In all disciplines NSF supports
the acquisition of computing equipment.
The Federal Government and Future Supercomputers
The recent history of the industry is one of progression in the development
of increasingly powerful general purpose machines. Class VII machines are
already under development with, for example, CRAY II planned to be available
in 1984-1985. It seems likely that this trend will continue and Class VIII
machines will appear in the early 1990's unless likely vendors are discouraged
from making the necessary investment in R and D. There is a limited market
worldwide and stiff competition expected from Japan. Decisions by individual
companies will be made on the basis of their internal assessment of the
likely market which will be influenced by Federal decisions. However,
industry planners are likely to discount agency plans that project specific
long term needs for Class VIII systems as being far too uncertain to provide
a basis for decision.
The options for government stimulation of this area range from maintaining
the vendors' confidence that a Federal market will continue in the next
decade, to direct support of the necessary R and D, to issuing an rfp
for Class VIII systems. The members of the panel did not appear to
support the last two options. There may be other immediate options that
fit the role of other agencies; however, NSF can help maintain a strong
U.S. position in the field by supporting a strong program of:
- basic research in universities in advanced computing concepts,
including the software, algorithms, architecture, subsystems
and components.
basic research in problems that require major computer systems;
- the training of scientists and engineers who will be familiar
with the concepts, design and use of major computer systems.
- assured access for researchers and graduate students to major
computer systems.
The NSF is likely to have a direct, although minor, influence on the
market through supporting the use of Class VIII systems. The basic
research supported by NSF underpins the development of systems beyond
Class VIII.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
 Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4
- 3 -
Access to Existing Class VI Computers
The computing needs for university research include access to supercomputers,
specialized computational equipment and array processors with supermini-
computers. This memorandum considers the rising need for university
researchers in a variety of disciplines to have access to up-to-date large
computers. That such a need exists is the opinion of NSF staff involved
with research areas requiring computation and is supported by the Lax report
and other studies specific to certain disciplines. Opportunities already
exist for researchers in some disciplines to access Class VI machines, for
example, through NCAR and the MFE network. Additionally, several universities
and some industrial concerns have either acquired or plan to acquire CRAY I
or Cyber 205 machines and will sell time. Federal laboratories also have
time available for outside use.
NSF recommends that an interagency working group be established, under the
FCCSET, to examine options for meetinq an aggregated demand for access to
existing Class VI machines. The group will need inputs from experts in
industry and the universities. The options include:
- providing access to machines in Federal laboratories;
- providing access to existing university based Class VI machines;
- the purchase of commercial computing services;
- the establishment of a dedicated network or center.
NSF recognizes that it will have a significant role in this planning
process and in the implementation of recommended approaches to solving
the problem.
Coordi nation
Requirements for coordination through more permanent working groups, or
at the policy level, will not be clear until the FCCSET supercomputer
panel has reached firm conclusions on the first two issues addressed in
your memo.
However, in addition to a group charged with examining the access question,
NSF believes that coordination of some specific elements of computer R and D,
e.g., computer architecture may be important. NSF does not perceive a need
to strengthen coordination of broad research programs in computer science
and engineering which is already active and effective on an informal basis.
Sanitized Copy Approved for Release 2010/11/12 : CIA-RDP85M00158RO01000060021-4