WORKING GROUP INFORMATION

• Document Type: CREST
• Collection: General CIA Records
• Document Number (FOIA) /ESDN (CREST): CIA-RDP84B00890R000400060009-1
• Release Decision: RIPPUB
• Original Classification: S
• Document Page Count: 36
• Document Creation Date: December 14, 2016
• Document Release Date: July 7, 2003
• Sequence Number: 9
• Publication Date: November 2, 1981
• Content Type: MF

Approved For Release 2003/08/13 : CIA-RDP84B00890R000400060009-1
SR I. ry 11 ~h_ I "ER
PAP
2 November 1981
25X1
MEMORANDUM FOR: Members, Technical and Scientific
Facilities Working Group
Chairman
SUBJECT: Working Group Information
1. Background
As you are aware, the IHSA has requested your partici-
pation in a 2-3 week effort to help them identify the direction
the Agency should pursue in the areas of Technical and
Scientific Facilities that will be required in the '85-'89
timeframe.
The IHSA has been directed to develop a strategic plan
for the Agency's Information Handling systems by the end of
August 1982. To accomplish this task within the time con-
straints, they have devised a four phase approach. In the
first phase, the current one, a series of user oriented working
groups will discuss and clarify the goals or objectives of
Information Handling from their perspectives. The second
phase will require other working groups consisting primarily
of IH providers to address how the goals identified in the
first phase may be implemented within technical, budgetary
and other resource realities. The last two phases are concerned
with preparing and coordinating the draft and final versions
of the Strategic Plan. A schedule for the Strategic Plan
development is attached for your information.
2. IHSA Point Papers
The IHSA has prepared three discussion papers that can-be
used as a basis for focusing our views and structure our prod-
uct. In addition to the background information provided there
are a number of specific questions which will require our
Approved For Release 2003/08/13: CIA-RDP84BOCI
25X1
 Approved For Release 2003/08/13 : CIA-RDP84B00890R000400060009-1
response. Because of the limited amount of time within this
phase you are requested to give as much thought as reasonably
possible to these questions and to the formulation of goals
prior to our gathering on 9 November. Any initial quantifi-
cation of facilities required or documentation supporting
specific contentions will be appreciated.
3. Some Guidelines
a. The time frame for goal implementation
is 1985 through 1989.
b. Our focus should be on required facilities
and goals, not on solutions or implementation
methods.
c. The fundamental thrust of our pursuit is
to ascertain (as specifically as possible) the
technical and scientific capabilities we require
for our efficient functioning during the target
time period.
d. Recommend specific changes to existing
systems/capabilities are not our concern.
e. It is not fruitful in this phase of the
effort to pursue "general goodness" concerns,
such as commonality and interoperability, because
these are aspects of IHS implementation planning.
They will be major concerns at that time.
f. In addition to the questions and goals
posed, any additional topics which you believe
should be addressed in this area will be welcomed.
4. I look forward to an informative and productive
association. If you have any questions or observations con-
cerning this matter before our meeting on November 9th, please
contact me
25X1
Attachments:
As Stated
Approved For Release 2003/08113 : CIA-RDP84B00890R000400060009-1
SECRET
 Approved For Release 2003/08/13-:-ClA=RDP84B00890R000400060009-1
AGENDA
Technical and Scientific Facilities
Monday, 9 November 1981
Room 2E-29
0900 - Welcome and Introduction
Working Group Chairman,
NFAC 25X1
0915 - Discussion of the Computer Aided Design and
Interactive Graphics Point Paper. Comment
and Review
1030 - Presentation on Economic Modeling
Issues
and Modeling
1300 - Discussion of Modeling. and Mathematical Analysis
Point Paper. Comment and Review
1430 - Presentation on Array Processor Assessment
1530 - Discussion of Special Machinery Point Paper.
Comment and Review
1630 - Adjourn
Tuesday, 10 November 1981
Room 4F-31
0945 - Discussion of Tentative Goals and Objectives.
^Instructions for Working Group
Approved For Release 2003/08/13 : CIA-RDP84B00890R000400060009-1
SECRET
 Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
 Approved For Release 2003/08/13 : CIA-RDP84B00890R000400060009-1
PLAN FOR IHSs STRATEGIC PLAN
TASK
Sep
Oct
Nov
Dec
Jan
Feb
Mar
A r
May
Jun
Jul
Aug
Phase I: Objectives Def.
Working Group Session (Phased)
Synthesis
Report to Senior Mgt.
Phase II: Implementation Planning
Dev. of Planning Guidance
Planning (Parallel)
Phase III: Dev. of Integrated Plan
Dev. of Rough Draft Strategic Plan
Report to Senior Mgt.
Phase IV: Reconciliation
r"]
V
Reconciliation with Budget
Dev. of Final Report
Report to Senior Mgt.
Legend
17 Presentation
Approved For Re p j*P $0,13 : CIA-RDP84B0089OR000400060009-1
 -----"-Approved For-Release 2003108/13:-CtA=RDP84B0089OR000400060009-1
SECRET
I. Background
In this era of increased automation, the Agency increasingly
relies on "computerized" methods for meeting.its objectives. This is
especially apparent in the areas which recieve high visibility due to
commercial advances and areas which are particularly germane to the
Agency's mission. Office automation, information dissemination, and
security all come to mind and we see general functions or systems like
"Word Processors" and SAFE systems being used and/or implemented.
However, the areas of specific user needs, while often less
visible, are also impacted by technical advances. The Scientific and
Technical Facilities within the Agency which are available to users
must be assessed as to their adequacy and use. If we are to take
advantage of these advances in our analytical process we must be aware
of them and define our objectives and goals in using them.
II. Scope
This area has been broken down into three subjects:
-- Computer Aided Design and Interactive Graphics
-- Modeling and Mathematical Analysis
-- Special Machinery
Although there is not a clear break between them, they are believed to
encompass a large portion of the functionalities perceived to be
needed to support the Agency's analytical mission. Additional topics
may be included if they are deemed relevant.
III. Approach
Attached are point papers covering each area. They are intended
to provide background information as well as to focus on issues
pertinent to the Agency. These papers will be used as "talking
papers" in the working group and it is hoped they will elicit the
views of the various users of these facilities.
Within each paper are relatively specific questions concerning
each area. To the extent possible the questions are intended to
quantify specific user needs in the specific areas. It should be
emphasized that this need assessment is not a commitment and does not
guarantee a capability. The goals and objectives we have to establish
will be used as general criteria toward which the Agency can proceed.
Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
SECRET
 Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1 - -
SECT
Attached also are copies of the description of a high speed
parallel processor and the description of a parallel processing
technique used for graphic purposes. These are included for your
information with respect to the subject matter and to illustrate that
these areas are not trivial matters. Needs for these facilities will
,dictate that the Agency acquire and maintain the expertise necessary
to efficiently implement them as we move from the "office automation"
implementation era into the era of sophisticated technical processing.
IV. Top Level Questions
The governing question for this issue area is the magnitude and
character of the technical processing requirement foreseen for the
1985-1989 time period. If it grows as significantly as the internal
needs and external technological advances indicate it might, then
accommodating it is going to require a sharply focused response. Such
a response will involve investment, in new types of hardware,
architectural innovation, new software, and special personnel
resources. The implementation issues are complex, and would be
addressed in the next phase of this strategic planning.
The more specific aspects of this concern are:
-- What scientific and mathematical facilities do the Agency
require to meet analytical needs in the 1985-1989 timeframe?
What is their scope, expected use, and relative level of
need?
-- Given that there is a need for specialized scientific and
mathematical facilities, what steps should be taken to acquire
the expertise to define, acquire, install and maintain them?
What is the funding, planning, timing, and environmental
requirements necessary to meet these needs?
-- In rough terms, what computer capabilities (power, speed,
systems, hardware, etc.) are required to meet these needs?
What qualitative aspects in terms of growth, size of user
community and changed capabilities are perceived?
Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
 Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
SECRET
W6RKING PAPER
Technical and Scientific Facilities
Computer-Aided Design and Interactive Graphics
I. Scope
Computer-Aided Design (CAD) is that field of computer application
which supports the analyst (user) by automating the creation, storage,
retrieval, display, and manipulation of variables within the design
process. For purposes of this paper this is limited to physical
variables, such as structure, equipment, and personnel, as opposed to
non-material entities such as money, data, and policy. The result of
CAD is normally a visual display or an input to some other process,
for example, assembly drawings for manufacture, or resource
allocations for project plans.
Interactive graphics comprises the computer-based capability to
graphically display the results of analysis and to manipulate them to
meet the user's needs. Included is the capability to present overlay
information on a map or chart base and standard graphical presentation
of information, such as bar and pie charts, graphs, and tabulations.
Often information that is presented is derived from files and/or data
bases which are then manipulated as a result of the graphically
displayed data. Data manipulation techniques that are used in
creating these displays are often complex (regression analysis, orbit
smoothing, etc.).
The Agency's requirement for a CAD capability is as yet undefined.
An effort has been initiated by ODP to examine CAD concepts and
functional requirements. That effort is expected to culminate in the
establishment of firm specific functional requirements for particular
CAD systems.
However, an overall Agency capability in which specific CAD
requirements can be implemented is not being planned. The fact that
the Agency does have functions that can be supported by a CAD
capability and the fact that advances are being made in CAD technology
dictate that we examine this field to determine the future environment
that is required.
Applications in these areas can be characterized into the
following general areas:
-- Interactive Graphics
This capability provides the user with the facility
to display statistical information graphically (graphs,
Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
SECRET
 Approved For Release 2003/08/13: CIA-RDP84B0089OR000400060009-1
t
charts, tables, etc.). A VDT (or VDU) interface permits rapid
manipulation of these displays. Often the source of
data for input to these displays is a file or data
base of specific information.
-- Building Description/Maintenance
Systems of this type aid in the physical planning
process including space layout, utility movement
(power, water, etc.), engineering, etc. These systems
are useful in designing changes to existing structure
internals in that changes can be assessed as to impact,
cost, etc.
-- Cartographics
This aspect of CAD is concerned with the generation,
storage and use of map information. It can be used
to study/manipulate physical and political entities
and is often used in conjunction with other variables
and factors, water, economics, etc.
-- Engineering
This field offers a wide range of CAD capabilities
in very diverse areas. Fundamentally, the computer
is employed in the design process aiding especially
in the spacial relationship of various components.
This can be used in publication planning, individual
component design, chip design, etc. These techniques
can also be used in the "backward engineering" of
systems, processes, and components to ascertain their
particular constituents.
-- Process Control and Management
These systems permit the planning of moving processes
(traffic, heat, work, etc.). Further, they provide
for determination of supply definition, resource
allocation, capacity limits, etc. As such they can
be used to plan and follow production systems allowing
management to change and reallocate resources as needed.
Some commercial applications of CAD are now quite sophisticated.
In the heavy manufacturing area there are systems which provide
"complete" automation. These permit the user to design equipment
using a computer terminal and then use the results as input to the
manufacturing process where various equipment is "driven" by computer
commands. This is Computer Aided Manufacturing (CAM) and often the
Approved For Release 20x3108/13 : CIA-RDP84B0089OR000400060009-1
 Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
SEM.
two areas are referred to as one (CAD/CAM). Manufacturers in portions
of the shipbuilding and aircraft industries are currently using this
technique.
B. Agency Status
There is a limited application of CAD and interactive graphic
capabilities within the Agency at this time. Several interactive
graphics applications exist, notably the TACK and TAD systems. These
utilize map data bases and display pertinent information with them in
an interactive mode. Other specific graphic applications exist, among
them:
- RAMIS has a graphic filing and display capability, although
not interactive and many systems present graphic output
reporting.
- NPIC uses a CAD system developing some order of battle
information. Using observed data from imagery they use
computerized techniques to ascertain previously unknown
variables.
- OCO's Cartography division uses computers in designing
various products. These use computerized base maps
in some cases to provide the basis of other displays.
Graphical, text, and briefing materials are also pre
pared using the GENIGRAPHICS computer aided design
system, which will later be linked to VM as a data
source.
Some project oriented offices have used CAD in hardware
development to a limited extent.
There appears to be a great potential for use of CAD and
interactive graphics in the Agency's normal functioning. Interactive
graphics can be used in the analysis area in presenting the effects of
changes in dependent variables. Cartographic capabilities can also be
integrated into this process thus displaying visually the status of
and effects on political/ military/etc. situations.
CAD and interactive graphics present significant processing
requirements that are comparable to scientific and technical
processing. Even today, with our relatively minimal use of such
facilities we can see the effects of the unique machine loading that
they represent. Three-dimensional figures are manipulated quite
slowly on our general purpose data processing machinery. Two-
dimensional figures are enhanced, expanded, or rotated rather slowly.
Approved For Release 20037 : CIA-RDP84B00890R000400060009-1
SECC
 Approved For Release 2003/08/13: CIA-RDP84B0089OR000400060009-1
SECRET
The mathematical operations involved are chiefly large matrix
operations and fourier transforms. These are the types of vector
arithmetic that point towards array and parallel processors, just as
in modeling and scientific analyses. Thus a significant automated
graphics requirement is likely to push us towards special scientific
capabilities, just as a significant growth in modeling and scientific
analyses will. Attachment A, from the March 1981 issue of IEEE's
Spectrum, describes some of the current thinking with respect to
complex graphics implementation on array processors.
1. What is the Potential of CAD in Backward Engineering of Foreign
Systems and in Engineering of New Agency Systems?
Backward engineering is the process of determining a systems
design configuration and internal functioning from external
observations. CAD could provide the same assistance to backwards
engineering by analysts dealing with weapons systems, military
hardware, and production processes as it does to manufacturers in
developing and producing the system.
2. What is the Potential of Interactive Graphics to Support
Cartographic Functions?
Interactive graphics support could also be used'more extensively
in the publication production area. Maps, charts, and reports
figures, etc., today are largely produced using "cut and paste"
techniques. Publication production using completely automated
techniques is broadly applied in the commercial arena to reduce costs
and improve the quality of the product and should be pursued within
the Agency. Additionally, graphical representation can be used to
monitor the poduction process itself, thus giving management an
insight into, and better control of, each aspect of production.
3. What is the Potential of CAD to Support Resource Management
Functions in the Agency?
Computer support can also be used in the resource management area
to graphically depict relative needs and availability of expertise,
expendibles, and processing functions. PERT-based techniques are
integrated with resource allocation and scheduling functionalities in
the CAD environment. Their application, especially to rather complex
projects, would aid management in ascertaining, allocating and
scheduling resource needs.
In the area of physical design, implementation, and maintenance
many CAD tools exist. In the physical plant area the Agency could
benefit from a coordinated CAD system to track all aspects of the
physical composition of buildings. Security, maintenance and
logistical components would benefit from an integrated CAD system
Approved For Release 20031b8/13 : CIA-RDP84B0089OR000400060009-1
SECRET
 Approved For Release 2003/08/13: CIA-RDP84B0089OR000400060009-1
cSy: J
depicting the location of all power and communications lines, safes,
phones, people, etc. This could be a shared system between the
various components and would greatly facilitate the building
maintenance and change process.
We are experiencing the same problems that forced the application
of the CAD techniques in the commercial environment: inadequate
numbers of qualified people to deal with developing and updating
drawings in an increasingly complex environment and a records validity
problem associated with the compounding of the normal human error rate
in that same increasingly complex environment. The cost of obtaining
additional personnel to cope manually with the consequences of the
geometrically increasing complexity of the environment are greater
than automating. Automating, furthermore, postures the organization
to continue to cope with ever increasing complexity.
With respect to CAD and interactive graphic capabilities what are
the perceived functions which they could support and what are. their
respective relative importance?
- Support in the analytical process
Statistical presentations
Mapping displays
Relative force displays
Economic trending
- Production/Management Control
- Engineering Support (including backward engineering)
- Publication Design and Production
- Building Logistics
- Communications Systems
To what extent is a "general" CAD and/or interactive graphic
capability needed to support relatively small and unique applications?
What is the need for the Agency to acquire a body of expertise in
the CAD area?
Approved For Release 2003/09/13 : CIA-RDP84B0089OR000400060009-1
 Approved For- Release-200=8/13 :- CIA-RDP84B0089OR000400060009-1
SENT,
Technical and Scientific Facilities WORKING PAR
R
Modeling and Mathematical Analysis
I. Scope
The concern of the strategic planning for modeling and
mathematical analysis capabilities is the need for specific, required
functionalities to provide computer assistance to analysts in
assessing external phenomenon. The application of these tools is
expected to steadily and rapidly increase because they provide the
capability to integrate the effects of very large arrays of
observables or consider a multitude of "cases" describing a particular
entity, function, or system. When appropriate, they permit the
continuing cycling of cases necessary to arrive at a "best" solution.
Modeling can be used in analytical processes ranging from economic
and political to hardware and behavior. Of principal concern to this
assessment are the "larger" models, either in terms of input data or
complexity and scope of the analytic processes. Many of the large
models have significant on-line data base and applications software
storage implications.
Currently the Agency uses modeling techniques in many intelligence
production cycles. They are used primarily in NFAC by the various
production offices. By nature they are analytic support tools and
therefore usually tailored by a user to his specific needs. The NFAC
Five Year ADP Plan (20 May 1981) cites many current modeling "systems"
and "tools" among them:
These models cover a broad range of economies and
are used in the assessment of individual country
economies as well as regional assessments. Included
are models which follow energy production and use.
The TROLL system is a major tool in this environment.
Models here are used to assess social change,
attitudes, political instability, etc. Techniques
used include scaling, Bayesian analysis, cross-impact
studies, election forecasting, and social simulation
models.
Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
UPUT
 Approved For Release 2003/08/13 : CIA-RDP84B00890R000400060009=4
SECRET
- Strategic
Assessment of'Soviet and Chinese military costs
and expenditures are produced. Force strengths
are calculated and refined for trade-off studies
(NATO vs Warsaw Pact, for example).
- Systems
Modeling is used extensively in defining and
assessing particular systems and hardware. Aircraft,
ships, antennas, etc. are examined and particular
variables are used to determine others. These, in
general, are specific models created for a particular
device, system, or activity.
Natural and human caused phenomenon are studied
using models including crop production, meteorological
effects, transportation, fuels, and metals. The
recently developed Meteorological Agronomical
Geographic Analysis System (MAGAS) is used for many
applications in this area.
Except for laboratory environments, such as those that are found,
in facilities like OSO's signal analysis center, our modeling and
scientific analysis are done on general purpose computer machinery.
Primarily, this machinery and the system software that drives it are
designed to do data processing. Such machinery currently adequately
supports the need, but principally because of the relatively low level
of scientific processing compared to to the level of general data
processing.
Because the current machinery is designed principally for general
data processing, it suffers from several disadvantages in doing
scientific processing. One is that it has inadequate precision for
such functions as inverting very large matrices or integrating
equations to determine trajectories with great precision. As a
consequence, such machinery is normally used in a double precision
mode for such functions. Double precision typically slows down the
general purpose computer by significant factors. It is thus,
relatively inefficient in doing scientific processing. A second
problem is that the machinery is not designed for the basic
characteristics of scientific processing versus general data
processing. Scientific processing lends itself to a significant
degree of parallelism in processing. If advantage is taken of this
characteristic, much more efficient processing can be done.
Some hardware that is commercially available is' designed to
support scientific analyses, and does a much more efficient job than
general data processing machinery with existing applications packages.
Approved For Release 2003/08/43 : CIA-RDP84B0089OR000400060009-1
SECRET
 Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
SECRET
Principally this superior performance is achieved through greater
precision, which avoids the problems of double precision, and through
a basic parallelism. The latter is provided by such devices as
separate memory and data caches, and an ability to parallel I/O and
arithmetic processing.
To take advantage of array-processors, however, new software may
be required. Such software has to define vector operations so that
array processors can be directed to perform parallel operations. This
means that if we see ourselves moving in this direction there is much.
more required than just installing special machinery. New software is._
required, and the higher order language employed is a significant
factor in the effectiveness of the implementation. FORTRAN, for
example, supports parallel processing. The new DoD language, Ada, is,.
specifically designed to support parallel processing, which is a major
requirement in the embedded systems for which it was developed. This
language, and its support environment, are expected to be fully
debugged and mandated on U.S. Army systems beginning on 1 January
1983. This timing may be a bit optimistic, but reflects the commitment
to and support of this development by DoD. The most pessimistic
estimates of the availability of a "clean" Ada and its environment are
two years later.
There is a strong indication that the use of models, modeling
techniques, and mathematical analysis will increase in the future.
Developments in the private sector will provide more such tools and
methods and make them more widespread and understood.
The appreciation by analysts of the capabilities of analytic and
modeling tools is rapidly growing. Increasingly, they are found
essential to synthesizing a vast array of data and extracting from it
precious information concerning capabilities, problems, and even
intentions. What has frequently been a barrier to these applications
is the ability of the analyst to interface first with the IHS via a
VDT, and then his ability to understand and operate the model. With
the academic training that technical students now all receive in
computational techniques, the increasing presence of such trained
individuals on our staff, and the changing attitude by older analysts
towards such tools, this barrier is rapidly diminishing. So many
individual analytic successes are now associated with the application
of such tools that there is a general appreciation that their
exploitation is a big factor in superior job performance. This is an
individual operational factor that is going to be a powerful driver in
the increased applications of models in the future.
Modeling and mathematical analysis techniques can and do impact
computing resources greatly. Sometimes this impact is observed by the
fact that the resource demands of most models are generally relatively
Approved For Release 20038 /13 : CIA-RDP84B0089OR000400060009-1
 Approved For Release 20% 0$ cIA-RDP84B0089OR000400060009-1
light over the periodic accounting periods. When they are running,
however, their demands may be quite high. In fact, operating
restrictions are frequently placed against models because of their
potential to disrupt other IHS resource users. If the applications of
modeling and mathematical analysis are seen as growing significantly
in the planning period--particularly relative to data processing--then
specific architectural and organizational upgrades are indicated.
Issue 1. What are the specific areas in which modeling and
mathematical analysis may be used?
Specific areas in which these tools may be used are of interest.
Especially important is the definition of analytical areas wherein
they are not being significantly used today, but are likely to be
exploited in the future. This will allow the Agency to procure
functional capabilities of a general nature which can then be used for
specific applications.
Issue 2. What expertise is required to take advantage of these
techniques?
Generally, these are complex and detailed areas of discipline.
The time and effort required to understand and use them is great and
leads to specialization. Comments have been made by Agency experts
that is takes a very large amount of time to become fully
knowledgeable and capable in using some of the more sophisticated
models. Time periods in the range of one to three years have been
mentioned. Such time periods present major career and organizational
problems to the Agency. The Agency needs to figure out how to deal
with the expertise problem. What expertise do we need and for what,
how diffuse or concentrated should it be, and how do we make the
necessary concentration attractive in a career sense?
Issue 3. What problems are there in acquiring and using these special
techniques?
Models and analytical methods can present special problems in
their procurement, installation and operation. They may require a
special DBMS or language, or they may be developed to run on different
machinery than we have. They may require special expertise to operate
and maintain. An assessment of current problems in these areas is
needed in planning the future environment.
What changes/increases are envisioned in the areas of modeling and
mathematical analysis? Quantify, to the extent possible, the number
of models and mathematical analysis "systems" envisioned and quantify
the amount of computer resources required to meet these needs. (See
Table 1)
Approved For Release 2003/0113 : CIA-RDP84B0089OR000400060009-1
 Approved For Release- 200-3/08113: CIA-RDP84-B0089OR000400060009-1
SENT.
Have the costs in procuring these functionalities been
incorporated in other long range planning?
B. Analytic Tools and Techniques
What new/changed techniques are envisioned? Specifically, in what
analytical areas are changes envisioned and what new techniques do
they require? For example, is more emphasis needed in statistical
modeling vice trajectory analysis? Prioritize needs if possible.
What problems have been encountered in acquiring and using these
techniques in the past and what are envisioned in the future?
Specifically, cover procurement, installation and use problems.
Why does it take so long to develop a thorough understanding and
facility in some models; are we actually developing the.professional's
analytic skills to the level of the model? Are the models poorly
documented? Are the models poorly structured so that it is very
difficult to undertand them, even given a good understanding of the
theory? Do the models have unnecessarily complex input data and
control requirements or do they have too many options for the typical
Agency application?
What need exists for the management of model and mathematical
analysis capabilities--their procurement, maintenance and use?
Consider their sources, costs, and expected uses.
Approved For Release 2003/0J13 : CIA-RDP84B0089OR000400060009-1
SECRET
 Approved For Releae2D08/13 : CIA-RDP84B00890R000400060009-1
TABLE 1
Systems/Functions Impacted by Technical and Scientific Facilities
System/Function
Description/Comment
Telemetry Analysis Display System; Response time and precision
needed for Vector Arithmetic and Fast Fourier Transformation
SCAM II Soviet Cost Analysis Model; Larger data base and precision needed
TROLL General Economic Model; Increased size (to 10,000 equations)
and speed required; Possible use of Array Processor
CHALLENGE Oil Reservoir Model; Large core and CPU requirement; CRAY
vector processor used in development
MAGAS Manipulation of Meterological Data; Multiple Instruction
Multiple Data (MIMD) capability desired; Display capability
important
RADAR RADAR Signature Data Analysis; Tenfold throughput increase desired;
High precision required
TRAJ Trajectory Analysis; Higher throughput desired; High I/O rates
SOSAG Nuclear Weapons Simulation Model; High CPU use
Networking Models ORD projects - JAWS (anti-satellite model), Soviet Trans-
portation, Refinery, Hydrology, Agriculture, CW; CRAY-1
considered .
Hardware/Systems MVS, PRIME, TRACE, TAPEST, KADRE; Models to simulate/define
Models hardware/systems; Possible Backward Engineering need; Large
CPU potential
-6-
Approved For Release 2003/08/13 : CIA-RDP84B00890R000400060009-1
SECRET
 Approved For ReleasZ0GBP08/13 : CIA-RDP84B00890R000400060009-1
Systems/Functions Impacted by Technical and Scientific Facilities
System/Function Description/Comment
Energy Models needed to trace/simulate/follow world and regional
energy production and use
Economic
Warfare
Commodity
Specific country/region models required; often time-phased
tied to production cycles; See also TROLL
Several models exist; often tied to warfare types (conventional,
nuclear, naval, etc.)
Models assessing political situations, potentials; election
forecasting, threat analysis, stability assessment included
Modeling and tracing of grain, metals, drugs, etc.; includes
supply/demand analysis, status, production, etc.
CAD/CAM Design, graphical presentation, data manipulation, etc.;
specific areas of use.desired
-7-
Approved For ReleasT08/13 : CIA-RDP84B00890R000400060009-1
 proved-For``Rete-ase 2003/08/13: CIA-RDP84B00890R000400060009-1
SECRET
For our purposes, special machinery is considered to be of two
classes:
o Class I - General purpose computers (generally
mini's) dedicated to special applications, such
as signal processing.
o Class 2 - Machines which are not general purpose
but are in fact special machines optimized for a
specific functionality, such as array processors.
In following sections status information and planning issues are
presented.
An inventory conducted in March 81 determined that the
Agency had 71 minicomputers, most being used for specialized
applications. The count by directorate was:
DDS&T
DDA
NFAC
DDA uses their minicomputers primarily for information
systems related to logistics, personnel, and medical
applications; other uses include text editing and
composition, data entry, and training. The ODP
initiative to offer GIMS on a minicomputer is noteworthy;
this provides the flexibility for GIMS applications
to reside either on a central or dedicated facility
without conversion difficulties.
Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
SECRET
 Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
DDO uses their minicomputers for data entry, CRAFT,
language translation, and compartmented information
storage and retrieval.
NFAC applications include plotters, digitizers, library
automation, high speed text search, crisis management,
and data analysis.
The reasons for using dedicated facilities rather than
central computing facilities typically include:
o The special nature of the application
(e.g., signal analysis or text search)
o Extraordinary requirement for responsiveness
o Sensitivity of information
Acquisitions of dedicated computers are carefully evaluated
and controlled by the requesting organizations, OL, ODP,
and others. Their operation in performance of the dedicated
function for which they were acquired is not a strategic
planning concern.
The chief planning concern relative to such machinery is the
consequence of applications' growth vis-a-vis the limited power they
possess. As the demands for computing power of their resident
appplications expand, a classic problem is created. More powerful
machinery is needed, but the applications can only consume a portion
of the total the new resource'would make available. What is more, the
more powerful equipment usually requires a higher level of operational
support, and of a higher skill level. Cost effectiveness in resource
management thus points towards migration of the expanded application
to shared central facilities.
Machines of special functionality which are most frequently
discussed are data base machines and array or parallel
processors. The use of such special machinery could well
imply significant changes in the architecture of our current
centers. Thus, the value of the inclusion of such machinery
has to justify the required investment.
Data base machines have generally been of interest to
information service providers, moreso than to users.
Perhaps this is because the thrust of data base machines
is to move the basic data base management functions from the
current software implementation to a hardware implementation,
rather than provide new functionality at the user interface.
Approved For Release 20"Q8/13 : CIA-RDP84B0089OR000400060009-1
 is
Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
SECRET
The principal benefits to accrue to the user from data
base machines are improved performance and reliability.
One of the most likely performance payoffs is probably to
be derived through the greater efficiency in whole text
search. A significant improvement in this area might result
in easing or removal of existing constraints concerning
personnel and daily access to the functionality.
Data base machines have been a subject of research for
years and commercially available products are now beginning
to appear. Britton Lee Inc. offered one of the earlier
machines (IDM), a device which is being considered for
use in CAMS II. Storage Technology Inc. has also announced
a product (VSS).
ODP indicated in the Working Group on Information Handling
Facilities that they will be studying the applicability
of this technology. User supplied incentives to use the
technology should be in the form of data base needs
(numbers of data bases, sizes, and access rates), also an
input to Information Handling Facilities Working Group'.
2. Array Processors
Throughout the NFAC Long Range Plan for ADP there are
references to increased computational requirements.
One office specifically identifies array processors
as a solution to their computational needs. Whether
or not array processor technology is an appropriate
solution is surfaced as an issue in the next section.
For our purpose, the term "array processor" means a single
peripheral processor attachable to a general-purpose host
computer so that the tandem combination provides a much higher
numerical calculation capability than the host alone.
An array processor might be viewed as an intermediate step
in providing a high performance computational capability;
the ultimate, and of course a much more expensive solution
would be a supercomputer of the CRAY or CDC 205 class. For
certain applications involving sophisticated computations
on large arrays, an array processor can be very cost
effective compared to doing the same computations on
a large general purpose machine.* Signal processing is
one application, and indeed the Agency is in the process
of installing a Floating Point (AP-120A) machine for
that application. It has also been suggested that the TADS
configuration might well be augmented with an array
processor to achieve the additional power needed.
Some of the large telemetry and modeling applications
within the Agency may also be suited to such machinery.
ODP has researched the need to a limited extent and
some perceived needs for an Array Processor do exist.
-3-
Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
SECRET
 Approved For Release 2003/08/13 : CIA-RDP84B00890R000400060009-1
Ul
The attached manufacturer's literature (Attachment B) provides
some of the best succinct, clear discussion of the different types of
vector processing machinery we have seen. It provides a discussion of
the technological features involved and their implications. Hopefully
it will provide some insight into the complexity of issues relevant to
the planning that will have to be resolved should there be indicated a
need for such machinery. There are likely to be substantial
organizational and programmatical impacts to user organizations in
addition to the obvious architectural concerns relevant to their
inclusion by service providers. For those interested in investigating
the characteristics of array processors in greater depth, the
September 1981 issue of the IEEE's Computer magazine is dedicated to
the subject.
III. Issues and Questions
1. Does the projected growth in the modeling and mathematical
analysis environment point to a likely need for special scientific
machinery?
There are two principal factors which would indicate a future need
for special scientific machinery, such as array or parallel
processors: the quantity of the work and the unique demands of the
work.
The quantity assessment derives from growth projections relative
to existing models and functionalities currently resident on the
central systems, for example the TROLL model and TADS and from the
migration of processing from laboratory environments, such as OSO's,
into a centralized environment. The latter occurs as the nature of
the required processing becomes more routine, more production-like,
and the power of the required processor increases. When the required
power increases it is usually true that single applications can only
.partially utilize the more powerful machinery. For acceptable cost
effectiveness, the environment then becomes one of shared machinery
usage in a centralized environment.
The second factor, the unique demands of the work, derives from
requirements for special machinery capabilities to handle the intended
processing. For example, NFAC's TROLL economic model can currently
handle up to 1800 simultaneous equations. NFAC has determined that it
needs to expand this capability to be able to handle approximately
10,000 simultaneous equations. Such an order of magnitude increase
creates signficant computational problems. The number of operations
in matrix inversion, for example, goes up geometrically with the
number of equations, as does the required precision. Such
requirements may exceed the capabilities of available data processing
machinery, forcing the acquisition of special machinery.
Approved For Release 29$3/08/13 : CIA-RDP84B00890R000400060009-1
 Approved For Release 2003/08/13? CIA=RDP84BOO89OR00041J0060009-1
SECRET
2. How great a problem is the Agency?s thin strength in scientific
computation professionals in the central environment?
Today almost all of the development and maintenance on scientific
software is done by contractors. Development and maintenance
requiring special access is frequently done on the same Agency
equipment that is being used for analysis production, e.g., TADS and
some TACK applications..
One consequence of the dependency on contractors, and the lack of
a separate development environment forced by funding limitations, is
that we are unable to run very sensitive data on the machines because
of unavoidable contractor access. Granted that we would always like
to have a greater depth of scientific computation expertise, the issue
is the priority for it.
3. Is there adequate justification for the major software and
programmatic undertakings associated with development of a
scientific computation environment?
The implications of a special scientific environment go well
beyond special processors. Although array processors, as opposed to
parallel processors, can process existing FORTRAN code, taking full
advantage of such capabilities implies special software. Even in
FORTRAN, the software should be specifically written to execute vector
processing. For a major program in this area, shifting to languages
specifically designed to support such functions, such as Ada, is
probably indicated.
Because of the specialized skills required to operate, adapt, and
enhance software in this area, there would have to be a cadre of
Agency professionals with expert knowledge. This would not be an
insignificant personnel requirement on the part of both user and
provider organizations.
With a significant workload requiring scientific-type processing
security concerns might well dictate the need for separate production
and development environments on a continuing basis. The former would
be Agency-only; the contractors would have access to the development
environment, as well as Agency personnel. This split would make it
possible to run extremely sensitive data, now precluded where such a
split does not exist.
Approved For Release MF : CIA-RDP84B0089OR000400060009-1
 Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
APPLICATIONS
Computers
Attachment A
fast graphics
use parallel techniques
Designers of computer graphics systems exploit parallel processing
to provide the speed needed for interactive performance
The main strength of computer graphics is in its ability to exploit
the massive parallel processing capacity of human vision-the
capacity to perceive almost instantly complex visual patterns.
However, until recently, the graphics displays have generally
been prepared serially by the computer hardware. Because of the
large volume of information to be computed in three-dimension-
al graphics, this has meant long processing times, a problem not
critical for scientific and entertainment applications, but serious
for the interactive, real-time systems used in computer-assisted
design and in real-time simulations. In these systems, the main
trend now is toward the development of parallel processing hard-
ware that can dramatically decrease processing times.
Some large computer graphics systems using parallel proces-
sors are already operating, and many others are being developed.
While parallel processing systems arp currently limited to highly
expensive simulators, VLSI designs under development could
conceivably bring the cost of sophisticated computer graphics
down to the price range of personal computers. Given the rapidly
growing demand for interactive graphics and the suitability of
VLSI circuits for parallel processing of increasing sophistication,
it seems likely that this field will become one of the first to use
VLSI technology in a big way. (See "The technologist's own
'super computer'," Spectrum, September 1980, pp. 48-51.)
At the same time, new methods are being developed to enter
rapidly the large amounts of data in interactive graphics, and
these methods, combined with faster hardware, will enlarge the
already widespread applications of computer-assisted design.
Complex scenes broken into sections
When used in computer graphics, most parallel processing sys-
tems handle complex scenes or images by breaking them into sec-
tions. Each section, with its many similar calculations, can be
generated in isolation from every other section. In this way
special-purpose parallel processors, each handling only a part of
the final image, can work far faster than serial processors.
The greatest premium on speed of image generation is in real-
time simulations, where extremely complex images must be up-
dated 30 times a second in response to the actions of the user.
Such systems are used for the training of jet pilots and other mili-
tary personnel. It is not surprising therefore that such simulation
systems have been among the first computer graphics systems to
use parallel processing.
An example is the Computrol system developed by Advanced
Technology Systems of Fairlawn, N.J. The system can produce
30 full-color images a second, each using up to 30 000 edges or
10 000 light points. Many moving objects can be displayed, while
fog, clouds, textures, and transparent objects can be modeled
Eric J. Lerner Contributing Editor
realistically (Fig. 1). The system permits fairly rapid generation
of new data bases-a new airport can be programmed in a few
man-days, for example. Computrol is to be used in the F-18
Weapons Tactics Trainer, being built by Hughes Aircraft for the
Navy for operation in 1982. It will simulate maneuvering aircraft,
terrain, gunfire, and missiles and will be equipped to train two
pilots simultaneously. The pilots will be able to maneuver against
each other in simulated missions. Similar systems are also used
for training the crews of tanks, ships, and commercial aircraft.
The detailed architecture of the Computrol system is treated as
confidential by Advanced Technology Systems-in fact, most
current simulator designs are kept confidential, a practice that
has hindered progress in this field. However, the general design
of the system has been published, and it gives a good idea of the
concepts applied.
The Computrol hardware consists of eight subsystems or
blocks.. An off-line terminal is used for creating the "world"
within which the simulation operates and is connected with a
minicomputer that controls the modeling process. A conven-
tional CPU and its associated main memory contain the data
base for the simulated world and control the movement of ob-
jects through it in three dimensions. Three specialized units are
concerned with converting the three-dimensional world into a
two-dimensional graphic representation. on a CRT screen. The
frame processor projects the three-dimensional world into the
appropriate two-dimensional field of view and simultaneously
converts the objects into edge-based descriptions-that is, the
edges define the borders between differently colored patches.
The raster processor calculates the intercepts between these edges
and each scanline on the CRT. Finally, the pixel processor takes
the intercepts of the visible edges, together with shading data,
and generates the color and intensity of each pixel on the
scanline.
The main parallel processing features are in specialized proces-
sors. The frame processor uses parallel arithmetic units to per-
form the calculations that transform world coordinates in the
data base to eye coordinates centered on the apparent viewpoint
of the trainee. Similarly, the raster processor has parallel circuits
to calculate the intercepts, and each subsection of the pixel pro-
cessor does identical calculations for a CRT subsection.
Applications to CAD
While the current application of parallel processing to com-
puter graphics is mainly limited to simulations, the same tech-
niques would be of great use in computer-assisted design if the
hardware could be made sufficiently cheap. This has become an
increasingly urgent necessity, since CAD systems are now using
three-dimensional techniques that, with existing hardware, tend
to slow processing radically and prevent easy interaction with the
user.
34 Approved For Release 209BM&t2iio > Q 2000400060009-1
IEEE spectrum MARCH 1981
 Until recently computer-assisted design has beeen applied
mostly in the two-dimensional world of electronic design, but
now it is expanding rapidly into three-dimensional applications
in mechanical engineering and architecture. A typical commer-
cial package, Synthavision, developed by the Mathematical Ap-
plications Group Inc. of Elmsford, N.Y., gives the user the abili-
ty to create any arbitrary solid, to manipulate it and view it from
any angle, and to obtain its volume, weight, center of gravity,
moments of inertia, and other geometrical characteristics.
General Electric has begun using Synthavision in the design of
mechanical components, such as gear trains. Similar systems are
being used by the Oak Ridge National Laboratory in Tennessee
and the Lawrence Livermore National Laboratory in California
in the design of complex magnets for controlled fusion ex-
periments. Synthavision is also used in computer animation ap-
plications. In some cases, the computer-assisted design of com-
ponents has been supplemented by computer movie simulation
of what happens to the components under stress, thus auto-
mating both the design and test-evaluation procedures.
Synthavision and most similar systems use built-up complex
solids by using a set of primitive shapes, such as cylinders,
spheres, cubes, and wedges, as well as arbitrary shapes that can
be parametrically specified. The solids appear on the CRT screen
as if illuminated from a specified angle, and the user can adjust
the reflective characteristic of the object's surface to simulate dif-
fuse or specular reflectance.
Such techniques are also coming into use in construction
[11 Complex scenes like this are generated by a flier training
simulation system called Computrol, developed by Advanced
Technology Systems. Computrol uses custom-designed parallel-
processing hardware to produce 30 frames of three-dimensional
simulation a second. The system costs approximately $2.5 million.
engineering and landscaping. A group at the University of
Massachusetts has developed a program called Ecosite that
enables the user to construct land forms, to be used to reform
surfaces that have been disrupted by strip coal mining. The
system allows a designer to create landfill shapes that will blend
naturally into the surrounding topography.
These existing systems, now implemented on conventional
computers, would benefit enormously from the higher speed and
interactive modes that would become available with the.perfec-
tion of parallel processing systems now being designed.
A variety of methods of using parallel processors are under
development to speed various aspects of computer graphics
generation. Much of the work is focused on the hidden surface
problem: the elimination of those portions of objects that are
obscured by other objects. One particular software approach to
this problem, the Z-buffer method, is especially suited to im-
plementation by parallel processors, since the determination of
what surface is visible is done independently for each pixel. The
Z-buffer (described in "The computer graphics revolution,"
Spectrum, February 1981, pp. 35-39) is a buffer for each pixel of
the image that allows only the nearest-object pixel to be entered.
One architecture proposed by Frederic Parke of Case Western
Reserve University implements the Z-buffer method by splitting
the image into regions and feeding the calculated surfaces in each
region to separate parallel processors. Each parallel processor
then determines the appropriate intensity and color values for
each pixel in its area, loading them into the appropriate Z-buffers
as it does so. Since only the closest pixels in each raster point will
be allowed into the Z-buffer, the resulting image will auto-
matically show only the surfaces that should be visible.
This system has a few limitations. Like all Z-buffer systems, it
has difficulty in dealing with "aliasing"-the tendency of com-
puter graphics systems to turn diagonal lines into staircases
Approved For Release 2003/08/13 : CIA-RDP84B00890R000400060009-1
 Approved For Release 2003%08/13 : CIA-RDP84BOO89OR000400060009-1
because of the finite dimensions of pixels. Also, the system
'becomes inefficient if all the objects are concentrated in a few
?regions, because most of the processors will then be idle.
A second architecture, developed by Henry Fuchs of the Uni-
versity of North Carolina, uses a central broadcast controller to
distribute the.data on each object to each processor. The data is
broken up not by contiguous regions, but according to an inter-
lace pattern, so that each processor handles pixels scattered over
the whole of the image. This eliminates the problem of having
some processors idle if the objects are concentrated in a certain
area. However, this architecture turns out to be considerably
slower, in general, than the split regions approach.
Mr. Parke at Case Western Reserve has proposed a hybrid ar-
chitecture in which the input data is first split into a small number
of regions and then distributed within each region, as in the
broadcast approach. In this way the disadvantages of both ap-
proaches are minimized.
VLSI designs sought to cut costs
The best way to decrease the cost of parallel processing hard-
ware is through VLSI. This approach is being pursued by James
Clark and associates at Stanford University. He has designed,
and is in the process of fabricating, a highly parallel VLSI com-
puter graphics system consisting of a "geometry engine" and
smart image memory.
The geometry section of the processor consists of 12 identical
geometry engine chips, each containing about 55 000 transistors.
The processor performs the basic operations common to prac-
tically all computer graphics operations-transformations, clip-
ping, and scaling of two- and three-dimensional polygons. It can
perform about four million arithmetical operations a second,
processing 900 polygons or 3500 edges every 1/30th of a second.
Each chip has a basically simple-architecture, consisting of an
arithmetic logic unit, three registers, and a stack; all working
together to form four identical functional units. The 12-chip
system consists of 1344 copies of a single bit-slice layout.
In operation, the geometry unit first receives the coordinates
of polygons from a central processor and transforms them into
the coordinates centered on the viewer. Four of the chips per-
form this operation by a combination of 4 x 4 matrix multiplica-.
tions and vector dot products that accomplish the necessary rota-
tions, translations, and projections to place the polygons in their
(2) The geometry system developed by James Clark and col.
leagues at Stanford University uses parallel processing In a VLSI
architecture to carry out procedures common to nearly all com-
puter graphics. A scene consisting of lines, points, and polygons is
first rotated and translated to correspond to the viewing position
of the user. The scene is then "clipped" to eliminate those portions
outside the viewer's field of view. Then the scene Is scaled to fit
within the viewing area of a CRT screen. The resulting polygonal
coordinates are then passed to a smart Image memory.
Objects are Objects are Objects are scaled
obtained by central moved to to perspective
processing unit scene locations viewpoint and clipped
44
w
locations in image space (Fig. 2). Since each chip has four iden-
tical subunits, 16 multiplications are being performed simultane-
ously in parallel.
The transformed polygons are then passed to the clipping sub-
system, which determines what part of each polygon is within the
field of view. For the three-dimensional case, the field of view is
defined by five or six planes that bound the volume visible
through the viewport of the image space. Each chip clips the
polygons for one of the bounding planes and then passes it on to
the next chip. Each geometry engine compares the coordinates of
(3) The memory system for a 1024-by-1024-element display being
developed at Stanford University contains eight column Image
memory processors (IMPs) and 64 row IMPs that convert sections
of incoming polygons into alterations of specific pixels in the scan
lines of the output display (A). Each row IMP Is linked to one or
more 16.kb memories, and the system has an output of 160 million
bits per second. In operation (B), the edges of the sample triangle
shown are scan-converted by the column IMPs and the Interior pix-
els, by the row IMPs. The letters indicate which column IMP has
converted each pixel, and the numbers indicate the row IMPs.
- Column IMP B
? . Retrestt,
controller
the
the
ary,
to t.'
inte
it d
of t
mic
side
poi
mit
unl
sic
art
fa-
th,
co
Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
36
IEEE spectrum MARCH 1961
 the end points of the polygon edges with the plane equation of
the boundary surface. If both coordinates are outside the bound-
ary, the edge is rejected; if both are inside, the edge is passed on
to the next chip. If only one coordinate is inside-that is, the edge
intersects the boundary-the chip finds the point of intersection.
It does this by logarithmic search for the intersection point. Each
of the four subunits of the chip computes one coordinate of the
midpoint of the edge and determines if that midpoint falls out-
side or inside the boundary plane. If it falls outside, then the mid-
point of the line connecting the inside end point with the original
midpoint is then calculated and tested and the cycle repeated
until the desired precision of the intersection point is achieved.
Finally, once the clipping operation is completed, the dimen-
sions of the polygons are scaled to the size of the image viewport
-that is, the farther away the objects are, the more the boundary
area must be scaled down to the dimensions of the viewing screen
for correct perspective. Two chips-one for the x, y scaling and
the other for the z, or depth, scaling-perform the division of the
coordinates simultaneously and feed the finished results to the
smart image memory.
The second part of the computer graphics design is the image
memory, a high-performance system for scan conversion. This is
the process of determining which pixels on the screen correspond
to the calculated images. The image memory is composed of a
parent processor, an array of image memory. processors, or
IMPs, and a refresh controller. The IMP array consists of eight
column IMPs and 64 row IMPs, each of the lgtter being respon-
(4] Procedural modeling employs a set of rules to generate more
complex objects from simpler ones and combines both specific
data entry and computer-generated repetitions. An.example
graphic, developed by Charles Csuri and associates at Ohio State
University, illustrates how entire city blocks might appear.
sible for 16 000 pixels of a 1024 x 1024 array (Fig. 3). The IMPs
are connected with the CRT pixels through a two-level hierar-
chical busing structure, with interleaved processors along each
bus. The interleaving is such that for any 8 x 8 array of pixels on
the screen, each pixel is controlled by one processor. Thus, as in
the Fuch's broadcast scheme at the University of North
Carolina, each processor -controls pixels scattered across the
screen rather than concentrated in a single contiguous area. Each
of the IMPs is a single LSI chip that contains two main functions,
a linear difference engine and a memory interface processor.
In operation, the geometry engine passes the characteristics
and locations of the elementary polygons to the parent processor,
a standard microprogrammed chip. The parent processor pre-
pares the polygons for scan conversion and broadcasts the
resulting data to all of the column IMPs. The polygons are
represented by the coordinates of their vertices. Each column
IMP Q;-IMP) uses its linear difference engine function to
calculate what part of the line falls within the column controlled
by its row IMPs and sends this information to the R-IMPs that
cover that part of the column. The R-IMPS, in turn, use their
linear difference engines to calculate which individual pixels
should be altered and send this information to the memory inter-
face processor for storing. At regular intervals, the refresh con-
..roller sends a signal to each of the memory interface processors
to obtain updates of the new pixel values and uses them to form
the new image on the CRT.
The edges of the polygon are thus converted by the C-IMPs to
their new values, while the interior is converted by the R-IMPs.
This architecture is being modified to implement the shading and
coloring algorithms of most polygon systems. The modified ar-
chitecture will obtain the shading values for the interior of the
polygon by interpolating between the values for the edges. In ad-
 Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
dition, the system can be extended to the sort of broadcast
Z-buffer hidden surface remover described by Dr. Fuchs.
The Clark architecture shows a conceptual similarity to that of
the system developed by Advanced Technology Systems, with the
column IMPs performing similar functions to the raster proces-
sors and the row IMPs analogous to the pixel processors.
Alternative ideas explored
A number of alternative ideas being developed use parallel
processing to speed certain special functions useful in computer
graphics. One example is a two-dimensional shift register design-
ed by George Chaikin of the Goddard Space Flight Center office
in New York City and Carl Weiman of General Electric.
This device is intended to speed the calculations involved in
two common graphic transformations: scale changes, or "zoom-
ing," and rotation. In a conventional system, these transforma-
tions require many individual calculations to change the coor-
dinates of each pixel in the image. Instead, Mr. Weiman and Mr.
Chaikin have proposed using a hard-wired polar logarithmic
transformation data channel to convert rotation and scale trans-
formations into translation motions on a shift register. The data
channel would connect pixels in the image plane arranged in a
polar logarithmic pattern to those in the shift register arranged in
a rectangular pattern.
In other words, a circle in the image plane is always mapped in-
to a vertical line in the register, and a radius in the image plane is
mapped into a horizontal line in the register. Rotation of the im-
age is achieved when each pixel register is commanded to shift its
content to its neighbor above or below. Scale conversion is
achieved when each register is commanded to shift its content to
its neighbor on the left or right. A single command can thus per-
form the work of many coordinate calculations.
While parallel processing will markedly speed computer
graphics processing, the most efficient use of such savings in time
will necessitate faster methods of data entry. Complex design
problems in three dimensions often produce difficult problems
simply in getting the design concepts into the computer in the
first place. A number of techniques for data input and structur-
ing are both easing data entry and simultaneously making some
processing tasks more efficient.
Two of the most important techniques are the related ap-
proaches of procedural and hierarchical modeling. In procedural
modeling, a set of laws is used to generate more complex objects
from simpler ones. The hierarchical approach breaks down com-
plex objects into simpler components or simpler representations
with less detail.
Using procedural techniques, 'which combine both specific
data entry and computer-generated repetitions where necessary,
Charles Csuri and co-workers at Ohio State University . have
developed a system for the design of buildings by use of stand-
ardized components. Computer graphics can then be used to
"construct" entire city blocks of a variety of such standardized
buildings (Fig. 4). The results give one a realistic view of how the
buildings would look in a city. An entire downtown area of two
thousand buildings was designed with this system in less than two
weeks. Such modularized techniques may have important appli-
cations in West Europe, where modularized building is far more
common than in the United States. It was, in fact, such tech-
niques that made possible construction of the elaborate data
bases in systems such as Computrol.
Hierarchical techniques take into account the fact that as an
object becomes more distant, less detail appears, and thus it is a
waste of computing power to calculate distant objects to the
same precision as nearby ones. In a hierarchical data base, a
single object may be represented by a number of representatives,
each having greater detail than the previous ones and the more
detailed ones being used for when the object is closer to the
viewer. Hierarchical representations also speed such processes as
hidden surface elimination, since if it is found that an entire ob-
ject will be obscured in a given image, elaboration of that object
to finer degrees of detail will be obviated. Thus, once it is found
that one building lies entirely behind another, the exterior win-
dows, doors, and so on in the building will not be calculated at
all.
Degree of detail required can be varied according to circum-
stances. Thus, moving objects can be calculated in less detail
than motionless ones, or whole scenes can be less rigorously im-
aged if the field of view is moving rapidly. Another advantage of
hierarchical data sets is in the reduction of memory storage
requirements for graphics processors. A working set of images
can be formed, consisting only of the images that have, in recent
frames of the sequence, been resolvable. This working set can be
kept in a fast access memory and only slowly changed or replen-
ished as the field of view changes.
A third technique of importance in efficient data entry is the
use of piecewise continuous surfaces, or splines, for defining
continuously curved surfaces. In many CAD applications, the
manipulation of sculptured surfaces, such as ship hulls, is ex-
tremely important, yet with even the fastest processing capabili-
ties, point-by-point entry of such curves is very time-consuming.
Spline surfaces simplify the creation of such sculptured surfaces
on a computer graphics system.
A B-spline, a widely used type, consists of a network of points,
each having associated with it a set of vectors that define the
directions of curvature of the surface at the point. The combina-
tion of points and vectors can be used to produce smoothly curv-
ing surfaces that can have almost arbitrary characteristics. One
can modify the surfaces by selecting a given point and either
moving it or changing the associated vectors.
The combination of faster input algorithms and the increased
speed and decreased cost of parallel-processing hardware will
rapidly make very powerful CAD graphics systems widely avail-
able. In the next few years, such systems will be becoming a
standard tool in engineering.
For further reading
A description of the Computrol system is given by Sam Ranj-
baran and Ron Swallow in "Graphics of Complex Images in
Training," Second IEEE Workshop in Picture Data Description
and Management, 1980.
Frederic Parke describes two approaches to parallel processing
in "Simulation and Expected Performance Analysis of Multiple
Processor Z-Buffer Systems," SIGGRAPH1980, pp. 48-53.
James Clark's designs for VLSI computer graphics systems are
outlined in "A VLSI Geometry Processor for Graphics," Com-
puter, July 1980, pp. 59-69, and also in "Distributed Processing
in a High Performance Smart Image Memory," Lambda;
Fourth Quarter 1980, pp. 40-45.
Hierarchical data organization is discussed by Steven Rubin
and Turner Whiffed in "A 3-Dimensional Representation for
Fast Rendering of Complex Scenes," SIGGRAPH 1980, pp.
110-117.
Robert Marshall et a!, outline a system for procedural data
generation in "Procedure Models for Generating Three-
Dimensional Terrain," SIGGRAPH1980, pp. 154-159. The use
of B-splines is examined by David Rogers and Steven Satterfield
in "B-Spline Surfaces for Ship Hill Design," SIGGRAPH1980,
pp. 211-217.
38 Approved For Release 2003/08/13 : CIA-RDP84B0089OR000400060009-1
Lerne -Fast graphics use parallel techniques
 pproved 'Ftlase-7'8t1 CIA'kR84B00890R000400060009=9
The many capabilities of the::_:_,
HEP hardware.: are fully=
supported by. HEP System
Software so that the- poten-.
tial performance of the. sys-
tem. is realized. with- relative-
ease.- Using, the- available
-System Software,.. pro-
gramming HEP is very simi-
lar to. programming a con-`;
ventional system, and only.:
minimal:: additional; pro-
grammer -training is re-
quired
In addition to the obvious-,-:_
design goals" of fast.-;
throughput and the ability to;;:,
..solve very large and com
plex problems, HEP ist de-:
signed for ease of. operation,:-"
and to be highly effective.-_,,
across the full range of
general-purpose computing.
applications --
Tomorrows' Computer.';
Is Here? _,._Today
Denelcor's Heterogeneous-
Element Processor. (HEP) is
a large-scale (64-bit) high-
,:..speed digital:- computer-
other supercomputer ar-
chitecture.. obsolete. HEP
provides a totally new com-
puting environment. high-
speed, parallel processing of
heterogeneous-= data ele-
ments. HEP has been de-
signed for use in: scientific
and/or commercial applica-
tions which can. effectively
' utilize processing speeds of
ten million to 160.million. in-
structions per second. HEP
achieves this throughput be--
cause of its design which im-
plements the Multiple In
struction Stream Multiple
Data Str.eam? (MIMD) ar- .-..
chitectural concept :for the'.:,
first time in a. commercially:::.
:,available system.
HEP makes available to the
.user up to 1,024 indepen
-dent instruction streams or pro-
cesses, each with, its own data stream to
be used concurrently for use in; pro
gramming applications. This multiplicity
of instruction streams running in parallel -
-enables and encourages breaking. the application into its
component parts for parallel processing. Other features of
the HEP design provide the synchronization necessary to
facilitate cooperation between concurrent processes, and
eliminates the precedence delays which often occur when
parallel processing is attempted using more conventional
data. processing equipment An equal number. of Super-
visory Processes are available for processing the privileged
functions necessary to the support of the User Processes
for a total of 2,048 independent instruction streams.
HEP .hardware is modular
and field expandable:.
HEP achieves its high speed
performance through ad
vanced architectural. con-
cepts rather . than- through un
proven "leading edge technology" elec
tronic components. This provides the user
.benefits in economy and reliability.
HEP Parallel Fortran is designed for maximum similarity to
existing languages, with logical extensions as necessary to.
implement the advanced features of HEP.
HEP is designed for ease of maintenance in the event of
hardware malfunction. Maintainability features are an in-
tegral part of the hardware design, including an on-board
maintenance diagnostic system which implements an In-
teractive. Maintenance Language for diagnostic purposes.
Approved For Release 2003/08/13 CIA=RfPR4R00890R0004000R0009-1
 Approved .For- Release 200 ft08113' CIA'RDP8-4B00890R{)004000SOQ0f !
eta o eneous _iement
Evolution of Computer
Architecture
The earliest computers executed a single instruction at a Another approach to increasing the speed of computation
time, using a single piece of data.. The architecture of these was; to make multiple copies of portions of the SISD
machines, called S1SD (for Single Instruction,. Single Data. hardware. In this approach, called SIMD (for Single In-
Stream) computers, was straight-forward, and well suited struction, Multiple Data Stream),. the operand fetch,.;
to the technology of the times. As technology advanced execution and result store portions of the hardware. were
and computer users required greater performance, SISD replicated, so that the execution of a single instruction
machines were made faster and faster, using newer and
better components and designs. But a fundamental prob-:
caused several values to be fetched, computed upon and
the answers stored. For certain problems, this provided a
lem remained.. Although the execution of a. computer in-
struction is physically composed of several parts -instruc-.
tion fetch, operand fetch, execution and result store:- the .
SISD computer could only perform one of these at a time,
since each step depended on the completion of the previ-
ous one. Thus, three-fourths of the expensive hardware
stood idle at any given time, waiting for the rest of the.:7 .
hardware to finish operation. , .
SISD
Single Instruction,
Single Data Stream
substantial performance improvement.. With. sufficient.
hardware; entire vectors of numbers could be operated :
upon simultaneously.. However, as with- "look-ahead", SISD machines,. the occurence of test and branch instruc-
tions, among others, required'the machine to wait for the,
;..
total completion of the instruction before proceeding..The
test and branch itself could. make no use of the replicated
hardware.
SIMD
Single Instruction,
Multiple Data Strea
SISD designers attempted to.remedy this by a technique In addition, two new problems were created by the SIMD
called "look-ahead", in which instruction fetch for the next architecture. Substantial portions of most programs are
instruction was overlapped with some portion of the not vector-oriented. The computation of iteration variables
execution of the current instruction. This provided some and array subscripts is a scalar problem, for which SIMD
performance improvement. However, digital computer offers no speedup, and the collection of operands across.
programs, particularly those written in higher level lan- arrays is an addressing problem which many SIMD
guages, contain large numbers of test and branch instruc- architectures do not handle. As a. second problem, if an
tions, in which the choice of the next instruction depends SIMD computer has a fixed quantity of replicated execu-
on the outcome of the current instruction. In such cases, tion modules (adders; etc.), and if the length of the vector
"look-ahead" offers no speedup, and introduces substan which the user wishes to operate on differs from the vector
tial complexity to make sure that the partial execution of length of the machine, performance suffers and software
an incorrect next instruction does not contaminate the- complexity increases. The cost of computation remains
computation. Approved For Release 2003/08/13 :' Clk- R i dwwd bmeb" utilized
 -Approved For Release 2003/08113: CIA-RDP84B0089OR000400060009-1
r - ogeneous Element Process Us
1 I G L .,.
Den?iccr, C.
Evolution of Computer
Architecture, :
Continued difficulties with the implementation of high per-
formance, cost effective computation using single instruc-
tion machines have led to the development of a new
concept in computer architecture.
This concept, called MIMD (for Multiple Instruction, Multi-
ple Data Stream) architecture, achieves high performance
at low hardware cost by keeping all processor hardware
utilized executing multiple parallel programs simulta-
neously. For example, while an add is in progress for one
process, a multiply may be executing for another, a divide
for a third; or similar functions may be executing simulta-
neously, such as multiple adds or divides- In MIMD ar-
chitectures, cooperating programs are often called "pro-
cesses". Independent. programs may contain one or sev-
erai processes.
Process Process Process Process
1 2 3 4
processes. Since this arbitration of the state of memory
locations is handled by hardware and without affecting the
execution of unrelated instructions, the communication
delay: is short and the overhead is small.
MIMD computers may be used to execute either SISD or
SIMD programs. SISD programs are just MIMD programs
with no inter-program communication. Execution of mul
tiple identical MIMD programs is equivalent to execution of
an SIMD program.
In the SIMD case, MIMD computers may match the vector
lengths exactly, while using remaining resources for unre-
lated computation. Thus, high efficiency may.be main-
tained even through scalar portions of the code. But the
major, application of MIMD computers lies in problems of
Multiply
Add
Multiply
Because the multiple instructions executed concurrently
by an MIMD machine are independent of each other,
execution of one instruction does not influence the execu-
tion of other instructions and processing may be fully
parallel at all times.
Successful MIMD architectures (figure 3) also provide
low-overhead mechanisms for inter-process communica-
tion. In these architectures, data locations may contain not
only a value but a state. Processes may synchronize by
waiting for input locations to have the "full" state. Result
storage may wait for output locations to attain the "empty"
state resulting from f~{S~S1~8UIL rF~#rt l er 3fO 1i 3
sufficient complexity that straightforward vector computa-
tion is not feasible. In these cases, which include continu-
ous simulation and complicated partial differential equa-
tion solutions, MIMD architecture offers the only possible
method of achieving significant parallelism. Denelcor's
Heterogeneous Element Processor system is the only
commercially available MIMD computer.
 ' Approved For Release 2003/08/13: CIA-RDP84B00890R000400060009-1.
Diagnostic Maintenance.
Sub-System -: r
: Sub-Systeru':I
-----------
The I-IEP computer system consists of process execution
modules (PEMs), data memory modules and support pro-
cessors interconnected by a high-speed data. switch net-
work. All. data . memory modules are accessible by all
PEMs. Thus, processes executing in parallel in one or
several PEMs may cooperate by reading and writing
shared information in the data memories. Parallel pro-
cesses synchronize and pass information back and forth
using the full/empty attribute of each data memory loca-
tion. HEP instructions may automatically wait for an input
data memory location to be full before execution, and
leave the. location empty after execution. Instructions may
.also wait for an output location to be empty before execu-
tion and leave it full after execution. This communicati
ons
discipline allows processes to conveniently and unambigu-
ously pass information to other processes while executing.
The iull/empty attribute ensures that reads and writes of
inter-process variables will alternate and no information
will be lost. For locations used exclusively within a process,
the full/empty attribute is ignored and memory may be ac-
cessed conventionally.
Both normal and synchronized memory access are avail-
able to the Fortran programmer as well as the assembly
programmer.. Software modules in both Fortran and as-
sembler programs may be distributed across several PEMs
to achieve increased throughput. In general, design of a
parallel program is not affected by whether the program
will run in one or several PEMs.
HEP System
Structure
i (/O _ i9P:r~ Interfaces
Cache Peripherals
---------;--4- Sub-Systems
External -it
1 1/0
L-----------
DAC
Clocks
Discrete 1'O
ADC
In HEP, creation and termination of parallel processes in
an MIMD program is a hardware capability directly avail-
able to the programmer. Processes are created or termi-
nated in 100 nanoseconds by execution of a single HEP
instruction. Thus, processes may be created at any point in
a program where additional parallelism is required, and.
terminated as soon as their function is accomplished. Up to.
64 user processes may exist simultaneously within each
PEM in a HEP system..
In order to efficiently manipulate data, each PEM contains
2048 internal general purpose registers. PEMs auto-
matically detect and flag normal arithmetic errors (over
flow, underflow, etc.) and may generate traps on occur-
rence of these errors. Programs in a HEP system are-
protected from each other and relocated in memory by a
set of relocation/protection registers in each PEM. This
.allows multiprogramming in a HEP system with full isola-
tion of one user from the next.
All data and instruction words in a HEP are 64 bits long,
although PEM data memory reference instructions allow
partial word and byte addressing. The memory bandwidth
in a HEP system is 20 million words/second per PEM,
including the data switch network. Each PEM executes up
to 10 million instructions per second. The architecture of
the switch network.allows up to 128 memory modules of
up to one million words each and up to 16 PEM's. This
range of system configurations results in speeds up to 160
million instructions per second on 64 bit data and memory .
Approved For Release 2003/08/13 CI1k11*46639i0id@GtA060009-1
 Tocesso
Spare
Spare
HEP 'systems, may include high speed real-tim-e I/O. de-.-
"Vices connected to, the data switch: network ::These.devices
operate at memory speeds up. to 80 million bytes/second
Normal I/0-.devices' are connected to thee: HEP; system
,:through support` processors...::Thus?standard;-commercial 11 :: I/O devices and., controllers, may be , used for--.routine I/O
:functions. All standard 1/0 devices-.are- accessible through
ttie HEP operating system andFortrari;T[O
HE? Softwas!e
HEP systems support a batch operating system va . or
trap and assemblers programming aanguages. The; .HEP
.operating system-.provides input =and::: output;: spooling,
batch job scheduling, and full . op erator control of the sys
parallel capabtl esi I he t-ortranprogrammer has access to
all standard Fortran formatted and.,, unformatted .1/O
capabilities. In: addition to the relaxation of syntax com-
mon to many Fortran compilers, HEP Fortran provides the
programmer with the means for explicit parallel pro-
gramming. A math library is also available which generates
parallelism in the evaluation of known functions.,
The HEP Assembly Language allows the user to access all
of* the capabilities of the system in an efficient manner..
HEP Assembly Language subroutines may be included in
Process Execution
Module Structure
used sections of code. Assembly Language programs have
direct access to all' hardware capabilities, including. the
direct creation and, termination of arbitrary processes.
The. HEP Link-Editor binds programs and subroutines into
processes;.tasks, -and jobs. The input is from either HEP
Fortran or HEP. Assembler. The output is HEP machine
executable code which is input to. the HEP Loader- at
? _-execution time.::The,HEP- Link-Editor runs as a user job in
the HEP PEM
The HEP File System provides a large volume; high data
rate I/O capability via.the.HEP Switch to a HEP System
with multiple Process Execution Modules (PEM). Sequen-
tial access.to information stored in multiple moving-head
disk files. is provided to the system at data rates from 80 .
'megabytes .per second (the maximum input rate for the
switch), too- approximately 1 megabyte per second (the
rotating storage data rate). Random access to information
is provided with comparable bandwidth, depending on. the
logical file size and the access patterns.
The HEP Interactive Maintenance Language (IML) pro-
vides a sophisticated yet easy-to-use language for debug-
ging the HEP System. It is used in. conjunction with main-
tenance hardware, wither test slots in the HEP main frame
or off-line test fixtures.,.The language is procedure_'__
oriented, thus permitting complex. functions to be coded
into higher order procedures.
Approve&F -6r - *Release 200#081'1.3:: ct-RI P8dB00890-R0004O0.O60009-1
 Approved ; For. Release 2003/08/13: CIA-RDP84B0089OR000400.060009-1,:
H et~.r ogeneous Element Processor
The most obvious area of HEP application is the multi-
programming of ordinary SISD algorithms. This. applica-
tion. does not use the inter-process communications fea-
tures of HEP, but fully utilizes its computing capacity. Since
.'.HEP's parallel architecture allows more complete utiliza-
tion. of its hardware, the cost effectiveness of HEP multi-
programming is higher than for other machines of compa-
rable performance. Another benefit. of HEP's effectiveness
at conventional computation is that it can easily run all jobs
at a facility.- not just those which are sufficiently large or
important to be written in parallel.
The application for which HEP was originally designed was
the solution of systems of ordinary differential equations,
such as those describing flight dynamics problems. In these
problems, a substantial system of dissimilar equations must
be solved, often in real-time. Many of the functional rela-
tionships in the equations are empirically derived and must
be repetitively evaluated by multi-dimensional interpola-
tion in lookup tables. Historically, such problems could
. only be solved, with limited precision and great expense,
using analog computers. The HEP MIMD architecture. is
the first commercially available digital technology capable
of.effectively addressing these problems.
Another. application area well suited to HEP is the solution
of' partial differential equations describing continuous
media. These equations, which occur in fluid dynamics
-and heat transfer problems, are typically modelled using a
grid of lattice points within the continuous medium. The
",...behavior -at a point is a function of the values at its
-.,neighbors.. The HEP's architecture allows these problems
to be solved with full parallelism, even in the presence of
irregular or time-varying lattice geometry, or with complex
functional relationships between lattice points.
A fourth, and very general, application area for. HEP is that
class of problems for which a large number of discrete
.elements must be modeled or computed upon. Examples
of such problems are tree and table searches, multi-
particle physics problems, electric power distribution, and
fractional distillation simulations. In all cases, complex be-
havior at a number of sites must be modeled, and interac-
tion between the sites is critical to the result Such prob-
lems are easily solved on the HER
The computing requirements for each of these applications.
are different. To effectively supply the range of capabilities
needed, the HEP system is available in several configura-
tions: ,
Approved For Release 2003/08/13 : CIA-RDP84B00890R000400060009-1
 Approve 209 /9
t rogeneous Element Pro" Processor
HEP's building block. architecture-: offers total flexibility
enabling. the user to start with. the exact amount of compu-
ter power needed. As computing. requirements grow,.
HEP's . field-expandability allows, the user to easily and
"economically add hardware and software modules to ac---
the largest of applications. These advanced
features clearly place. HEP in the forefront. of. digital. com-
puter technology and provide strong competition for exist-
ing computer systems, both scalar and vector.
The evolution of HEP is a natural result of Denelcor's
on-going commitment to meet the market needs with
state-of-the-art, high-quality systems......
?10 Million to 160 Million Instructions per Second, Scalar-..,%
or Vector.
2,048 to 32,768 General Purpose, 64-bit Registers
262,000 to one Billion Bytes Memory Capacity
.Parallel Computing in Fortran
a Fail-Soft Architecture
?~Unbounded I/O Rates
Real-Time Synchronized.
Denelcor, Inc.
3115 East 40th Avenue
-Denver, Colorado 80205
303-399-5700
TWX: 910-931-2201
Approved For Release 2003/08/13: CIA=RDP84B00890R000400060009-1