Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 64
APPENDIX
Computer-Related Technologies
The computer-related technologies discussed in this report are applicable,
of course, not only to other large data storage and information systems
such as the one proposed for the Social Security Administration, but also
to other kinds of computer-based applications. The purpose of this
appendix is to pull together the discussion found in the body of the
report concerning the technologies of storage and memory, alternative
design concepts, data base, programming and software, and semiconductor
components, including integrated circuits and microprocessors. These last
components, especially, find diverse applications in the fields of
transportation, communications, and consumer appliances in the home.
Data Base Storage
The design of the Social Security Administration's future process,
as conceived by the agency, calls for the primary data base to be
stored at a central site--presumably the facility under construction
in Baltimore. This central data base would be connected to a group of
interactive terminals over a nationwide network. In evaluating this
concept, the panel considered alternative solutions for the storage of
data.
One alternative is the extensive distribution of both the process-
ing and storage of the information in the system's data base. The
distribution would be essentially to the district office level, with
1,200-1,500 data storage and processing sites.
Another alternative is a regional separation of the data base into
several large data storage and processing centers, a design concept
commonly referred to as regional. This would provide access to the
information through the same techniques as centralized design. The
major difference turns on the storage points, in this case from two to
six, each providing capability for the backup of other portions of the
data base.
Thus, the three alternatives described more fully in this Appendix
are the central, distributed, and regional design concepts.
64
OCR for page 65
65
Central Design Concept
,
The panel recognizes the advantages of the central design concept
in the following ways:
· The SSA's existing system incorporates a central design and,
therefore, the transition plan, involving personnel and
facilities, could be more readily adapted to this design than
to the other concepts.
.
The SSA is constructing a central data processing site in
Baltimore that could incorporate this data processing design.
· It permits more ready resource allocation toward modification
of the basic design and could be adapted more easily to changes
in legislative requirements or operating procedures than the
two other concepts.
Distributed Design Concept
The panel found the following significant shortcomings in the
distributed design concept:
It lacks the flexibility to accommodate easily any required
changes in data format, content, processing capability, or
operating procedures.
While the storage of the information could reasonably be accom-
modated at the local office where the information would be
used frequently, the mobility of the U.S. population makes it
likely that a significant percentage of the accesses to a
particular file will involve more than one district office.
Based on statistical information and the practical experience
of SSA field personnel, provided to the panel during the review,
the panel has concluded that there is considerable interchange
of information among various district offices. This would
require significant processing capability in each one.
The processing required to carry out the SSA process is substan-
tial. The data could either be processed at points within the
network or, alternatively, would have to be moved to a substan-
tive processing point on a periodic basis. [his would mean
either major processing capability at each distributed point, or
major communications capability between the distributed storage
points and significant processing sites, or both.
The integrity of this data base is, of course, uppermost in the
minds of the SSA planners. To maintain a high degree of
security, with uniform policies and procedures, would be more
difficult if the information were to be located at a large number
OCR for page 66
66
of storage points. The panel
recognizes that significant
security precautions are in place now and would always be in
effect at the local offices. However, at any time only a small
portion of the data base is located at a district office in the
present system--or in either of the proposed central or regional
design concepts. The degree of risk is less, therefore, with a
central or regional design. To provide for the security of the
distributed system would require that maximum security be main-
tained at all times at all district offices. While this would
not appear to be feasible from an economic standpoint, because
of the large number of district offices, it is feasible to
provide maximum security at a central site or even at several
regional centers. Furthermore, the access to a central or
regional data base from a district office could be regulated
with strict security procedures at all times, particularly
during hours when a district office is not manned.
.
The panel found that the potential for fraud against the SSA
organization would be greater with the distributed concept
because of the larger number of personnel who would have intimate
access to the processing capability and the storage media in the
district offices than the number with access in the central or
regional designs. In the latter alternatives, personnel having
access to the data and the processing capability would be more
restricted and could be more readily audited.
The Regional Design Concept
The panel has observed during its deliberations that the regional
design alternative should be considered for the future SSA process as it
evolves. As it has happened, the GAS has relaxed the fixed requirement
that the data-base be located in a single central site.
The regional concept is more expensive than the centralized design,
but a major consideration for holding this option open is the opportun-
ities for security and redundancy of the data base. Although it will
complicate the design to provide access from district offices to multiple
centers, the panel noted that this design offers great security for
information.
Data Base Structure
An examination of the basic structure of the SSA data base is a
necessary prelude to some of the fundamental considerations of the panel.
The data base is keyed primarily by the social security number (SSN),
and each client's record contains data that has an affinity for certain
functional requirements, such as the data that identifies the holder of
a particular SSN, the earnings data describing the client's earnings
history, and claims data, if, in fact, the client has made claims against
his social security account . Therefore, the data base may be viewed as
a two-dimensional matrix, as shown in Figures 6 through 9. SSN's are
OCR for page 67
OCR for page 69
OCR for page 70
OCR for page 71
OCR for page 72
OCR for page 73
OCR for page 74
OCR for page 76
OCR for page 77
OCR for page 78
OCR for page 79
OCR for page 80
OCR for page 81
OCR for page 82
OCR for page 83
OCR for page 84
OCR for page 85
OCR for page 86
Representative terms from entire chapter:
district offices
67
\Function
SSN \
f~xx-xxxx
Identity
Data
Earnings
Data
Claims
Data
Address
Data
_ XXX-XX-XXXX
E XXX-XX-XXXX
.,XXX-XX-XXXX
XX X-XX -XX X X
(XXX-XX -XXXX
XXX-XX-XXXX
E XXX-XX-XXXX
~n ~ XXX-XX-XXXX
~ X2
68
located on the vertical axis and the functional data categories on the
horizontal axis of the Figures.
The panel has attempted to itemize the advantages of the symmetric
and asymmetric approaches shown in Figures 6 and 7 respectively, to
segmentation of the data base.
The advantages of the asymmetric configuration are:
Improved capability to add new applications without disturbing
existing ones, inasmuch as the applications tend to have an
affinity primarily for a corresponding part of the data.
Capability to transfer applications from one processing site to
another, or if an application is shifted to another agency,
to disengage it from the SSA system.
O As a result of these two capabilities, the system would be better
able to respond to external priorities. An example of this might
be the implementation of the post-entitlement function initially
in the conversion process--one of the desired objectives of the
GAS plan.
Because the current data base is basically structured asymmet-
ricallY. conversion to an asymmetric configuration would probably
be simpler than to a symmetric conrlguraclon.
An asymmetric data base and configuration would offer great
potential to vary the hardware and software configurations. For
example, different central processing unit (CPU) architectures
could be used to support different applications. Such variability
is characteristic of the current SSA process in that there are
various CPU architectures supporting subsets of the current SSA
process.
Program distribution and maintenance could be simplified, because
the process for one function would be performed primarily in one
location, as opposed to the symmetric configuration where the
same processing would be done at all centers.
The one-site-per-function structure in the process would enable
better process specialization and possibly better utilization of
personnel by allowing them to specialize.
Improved capability would be offered, from a cost management
standpoint, to isolate the cost for each function.
O By segregating the functions with the asymmetric file, the con-
version and development process could be segmented into smaller
pieces.
69
The advantages of the symmetric segmentation of the data base are:
Easier load leveling by splitting the work symmetrically and
creating new centers as increased processing demand is identi-
fied.
Better capability for integration and synchronization of the
files and, as a result, better support of the '~whole-person''
design concept.
Simplified network routing because interrogation of the data
base in most cases is by specific account numbers or SSN's;
the network could route inquiries to the specific location
that maintains all the information needed for response.
Better traffic balance for the data communications network.
Less redundancy of data is offered in the data base. While
this is an advantage of symmetric segmentation, either symmetric
or asymmetric structuring of the data base in the future process
would offer less redundancy than is inherent in the current SSA
process.
Improved economics of the backup process would result, because
all of the centers would be similar. A need would exist for
only a generalized backup capability.
A single point of control at the central facility would result
in a more cohesive design.
Given these fundamental characterizations of symmetric and asym-
metric segmentation of the data base, the implications of the various
modes of executing this segmentation can be evaluated in some detail.
While the size of the data base suggests that segmentation is the best
format, the method used to effect it is critical to the fundamental
design of the system. Despite the attempt to keep the data base
independent of the application systems as a design objective, there will
always be a strong relationship between the applications processes and
the data structures. Therefore, the processing requirements are
important in making decisions on data base segmentation.
While the panel has concluded that the segmentation of the data
base does not need to be entirely symmetric or asymmetric, it has
observed that certain portions of the data base will have to be seg-
mented in an asymmetric format. For example, the identity data must be
maintained in a single center within the future process so that single
point assignment, control, and management of the identity data can be
maintained. Similarly, because of the nature of the earnings data that
are filed by the SSA, it may be desirable to maintain the data in an
asymmetric way.
70
If, however, the identity data were determined to be the only
information required to support the asymmetric format, then the primary
segmentation of the data base could be symmetric, with asymmetric sub-
segmentation, illustrated in Figure 8.
In contrast to the hybrid segmentation shown in Figure 8, the primary
segmentation for all of the data base could be asymmetric, with
symmetric subsegmentation depicted in Figure 9. The panel has concluded
that there might well be a hybrid segmentation similar to those shown
in Figures 8 and 9, with the data base segmented 'both symmetrically and
asymmetrically at various levels. For example, in handling claims, the
records that are referred to most frequently could be maintained on the
faster access storage media than those with a lower rate of recall. In
essence, media segmentation of the information within the data base
could be carried out in a symmetric subsegmentation within the primary
asymmetric segmentation of the data base.
The Whole-Person Concept
The design objective of the whole-person concept has been endorsed
by the panel as a desirable objective for the future SSA process. Adop-
tion of this concept will simplify operations and make it possible to
support the basically on-line data base to support the future service
objectives. While the symmetric segmentation of the data base appears
to align itself more readily with the whole-person concept, the data
base need not be segmented symmetrically in order to achieve this
objective. It is necessary, however, that the information be maintained
on-line in order to achieve an effective whole-person concept and to
provide the timely access that is a primary objective of the future
process.
Data Base Technologies
`- The panel has encouraged the use of standard technology and the
division of processes into discrete processing capabilities as a
necessity in the design of the data base. Future technology, including
storage devices, will continue to provide alternatives that will need
to be considered in the design effort to improve service levels and
reduce service costs. As technology improves, the design concept for
the direct on-line access to information should be able to accommodate
to the technological advances.
The same considerations holds for software development. The panel
also has suggested that, wherever possible, the data base design
incorporate software subsystems in wide use and maintained as standard
products. The reason for this is to minimize the need to maintain a
highly specialized technical staff for the support of specialized
software. The large size of the data base imposes some significant
limitations on what products are available from the commercial hardware
and software markets. As a result, some exceptions may have to 'be made.
The overall design of the future data base calls for an on-line
system. However, there will be variations in the speed of access to
71
\Function
SSN \
/ XXX-XX-X~XX
_ XXX-XX-XXXX
c XXX-XX-XXXX
c~
XXX-XX-XXXX
~ XXX-XX~-XXXX
/ X X X-XX-XX XX
/
XXX-XX-XXXX
c XXX-XX-XXXX
XXX-XX-XXXX
\
XXX-XX-XX XX
XXX-XX-XXXX
XXX-XX-XXXX
E XXX-XX-XXXX
a>
~n XXX-XX-XXXX
~, XXX-XX-XXXX
/ XX X-XX-XXX X
I XXX-XX-XXXX
c XXX-XX-XXXX
cn XXX-XX-XXXX
~ XXX-XX-XXXX
Claims
l~t:\
Address
Data
FIGURE 8 SSA Data Base - Hybrid Segmentation (Symmetric/Asymmetric)
\Function
SSN \
XX X-XX -XXX X
XX X-X X -XX X X
XX X-XX -XXXX
X X X-X X -XX XX
-
XXX-XX-XXXX
XXX-XX-XX XX
XXX-XX-XXXX
XXX-XX-XXXX
XX X-XX -XXXX
XXX-XX-XXXX
XXX-XX-XXXX
XXX-XX-XXXX
XXX-XX-XXXX
XXX-XX-XXXX
XX X-XX-XXX X
XXX-XX-XXXX
XXX-XX-XXXX
-
XXX-XX-XXXX
-
XXX-XX-XXXX
XXX-XX-XXXX
. I denti ty
Data
Segment 3 Segment 4
Data Da ta
: ::
.
~ ,
<;Pnm`!nt :)
FIGURE 9 SSA Data Base - Hybrid Segmentation (Asymmetric/Symmetric)
72
this information because of the economics of the storage media and the
options that are available for providing lower response speeds for
portions of the data base. Because of the size of the data base using
hierarchical storage, the ready availability of commercial hardware and
software systems may be somewhat limited. Accordingly, a cost tradeoff
will need to be made as to the degree of specialization required to'
support the data base. The programming and maintenance costs of
specialized support will need to be evaluated against any additional
hardware operating costs that arise from a generalized data base system
that is commercially available.
Stability of the data base is critical to the successful opera-lion
of the entire SSA process. Consideration is often given to the degree
of stability that is possible with specialized software as opposed to
software having more common usage and, therefore, a potentially higher
degree of stability.
Storage (Memory Technology)
_
memory:
At present, most computer systems include the following levels of
Archival (magnetic tape)
File (moving head disk)
Intermediate (drum or fixed head disk)
High-speed (core or semiconductor)
Buffer or cache memory associated with high-speed memory is not included
because it is invisible to the programmer.
For archival purposes, magnetic tape is satisfactory and will
continue to be used for this purpose during the foreseeable future.
File disks are both efficient and reliable, but a non-mechanical
replacement for them is desirable if a total reduction of
environmental restrictions on computer usage is to be achieved.
Paging from drums is very successful when used for batch appli-
cations. The extra amount of paging required for time-sharing
and the consequent need for a high level of multiprogramming requires
large high-speed memories. The development of such systems has been
hampered by problems of efficient and responsive operation under
varying conditions, and it is likely that a limit has been reached in
the complexity of operating systems. The replacement of the drum or
the fixed head disk by a device with similar characteristics, but with
greater speed, is necessary for future systems. Thus, the need exists
for bubble memories or charge-coupled memories, if they can be produced
at suitable levels of performance and cost.
A body of thought prevails that computer technology is at a point
at which drums or fixed head disks will disappear altogether. This will
happen as soon as the cost of high-speed semiconductor memory becomes
low enough that the economic advantage of providing the additional
capacity outweighs the complications to hardware and software.
73
Main Memory Technology
The same Very Large Scale Integration (VLSI) technology that will
bring about improvements in cost/performance in the CPU will also bring
about drastic improvements in the cost/performance of the main memory.
The problems of sharing the main memory among users as well as the
limitations on its size because of its cost probably will not arise in
a decade. Distributed processing will be encouraged by price and
performance improvements in main memory systems.
Auxiliary Storage
Much research is being performed on novel auxiliary storage
technologies, including magnetic bubbles, charge-coupled devices,
laser-holographic devices, and cryogenic devices.
Such research will gradually result in the introduction of new
types of auxiliary storage subsystems. Still, the potential for
improvement exists in today's conventional magnetic technology. Most
of the improvement will be in the form of increased area density of
recording--more tracks per inch horizontally across magnetic disk faces
and more bits per inch vertically. An area density improvement factor
of at least 40 appears theoretically possible. This will result in a
lower near-term cost-per-bit for magnetic disks than either bubble or
charge-coupled device (CCD) technologies can match. However, access
time to magnetic disks will remain a problem even if head-per-track
arrangements become general. For this reason, it is likely that a
variety of auxiliary storage devices will continue to be used through
at least 1985, with the newer technologies appearing first at the high-
speed, low capacity end of the spectrum, then gradually superseding
slower technologies as the costs drop of the former. Both magnetic
bubbles and charge-coupled devices could be in widespread use in, say,
~ .
rive years.
Summary of Storage Technologies
The SSA has limited its view of storage technologies to those
available today. Although this approach narrows the number of system
alternatives, it is practical unless some dramatic change, which is not
now foreseen, occurs in the period during which the system will be
implemented.
The SSA has correctly assessed the state-of-the-art in storage
technology, by assuming the continuing availability of a range of stor-
age devices based on relative cost/performance relationships. Large
capacity library stores are currently available with capacities of 472
gigabits, and access times of 15-30 milliseconds, at a cost of 30-50 cents
per megabit. Continuing developments in optics and magnetic media offer
the potential for significant cost and capacity improvements. In the
next five to ten years, potential improvements of three to five times
in area density with a two to three times reduction in cost are possible.
Similar improvements are possible in 1/2' tape, but the semi-automatic
74
operation of libraries and space economics will limit its potential
usefulness, except as an interchange medium for comparability purposes
or as a small system library. Optical storage technologies with read-
only/write-once capability are available, with the potential for future
density recording 100 to 300 times greater than magnetic media at
relative costs of 1 to 10 percent.
Intermediate storage products such as hard surface and flexible
magnetic disks, video disks, electron beam devices, bubble and
charge-coupled devices will be available for use in electronic libraries
as intermediate storage for on-line access. A number of factors are
stimulating the use of these devices:
· Low cost processors
· Diverse processor architectures
o Data sharing among several processors and applications
Increasing programming costs
Transition problems in changing several system elements
simultaneously
Hard surface non-removable disks are available with single spindle
capacities of 317 megabits (MB) at a cost of $1.81/MB/mo and access
time of 25 milliseconds. Improvements of four times in density and
access in the next decade with decreases of three to four times in cost
are potentially possible. The removable disks have capacities of 140 MB
at a cost of $5.56/MB/mo. Improvements similar in magnitude to those
on hard disks are potentially possible in a similar time frame.
Bubble and COD technologies are already being introduced to the
market. These will provide memories at a projected cost of about 30 to
40 millicents per bit in, say, ten years.
Computer Storage
Timely access to data elements (social security number, earnings,
etc.) is an important and persistent factor in the design of the SSA
system. In the past, programs have been written that access specific
records stored on tapes or disks in a unique sequence. When the size
of an individual data element is altered or new data elements are added,
major programming changes may be required. Frequently, the same data
elements may be required in differing groupings and/or different
sequences for other programs. The result is the need to synchronize
the updating of several different records on a controlled basis. The
SSA has recognized this problem for its current set of applications by
recommending the whole-person concept.
In order to accommodate changes in technology as well as modif-
ications to the system, both within existing and new programs, the panel
76
The same basic concept as above, but with two to six locations
providing redundancy of processing as well as permanent data
storage capability.
Distribution of the permanent data to the many district offices.
Because the potential exists for innovative approaches leading to
decreased on-line storage costs which can make feasible redundant storage
of data, analysis of potential design techniques always needs to be made
independently of the central processing complexes.
Programming
. _ .
Substantial progress is being made in improving the productivity
and effectiveness with which computers are being programmed. In addi-
tion, the dramatic decreases in computer hardware costs due to large
scale integration continue to make problem-solving by computer more
practical. As a result, the information processing industry has
clearly passed from being capital intensive to being people intensive.
The principal cost of using computers will continue to be that of the
people associated with these machines. Programming the computer and
planning its operations so that it is easy and economical to use will be
the technical and operational challenge of the future.
The most successful way of increasing computer programming
productivity is through high-level languages. Such languages enable the
computer user to perceive and solve the real applications problems with-
out getting entangled and confused with detail. Trends in high-level
languages are:
Building of special purpose high-level languages for data base
manipulation and for professional jargon language systems that
make them useful for unique applications.
Facilitating the structuring of the applications problems and
their solutions, so that the resultant programs are probably
correct when written.
Employing interactive graphics in both the definition of the
application and its programmatic solution.
Minimizing machine dependencies in the high-level languages.
Programming, in the meaning used here, is the art of describing
the problem to be solved in a form that results in effective computer
solutions to the problem, produced economically. Skilled people are
the essential ingredient in programming. System design is part of the
programming process in this context. One of the reasons that computer
programs have been so large and intractable historically is that the
programs were a people-built bridge between whatever hardware could be
obtained and an inflexible definition of the problem to be solved.
77
Programming is now evolving so that the system and hardware designs
must be done in conjunction with the coding of the designs in order to
achieve effective solutions. When the principles of top-down design
and high-level language programming are applied to the full power of
computers, communications and interactive display graphics, modern
information systems become flexible, easy to use and understand, and
economical to program and operate.
The rate of improvement in software technologies is appreciably
smaller than that in hardware. In each of the last three decades,
programmer or software effectiveness has increased by only a factor of
about five. By contrast, for each of the decades, computer hardware
as measured by cost/performance has improved by at least a factor of 20.
These relative rates of improvement probably will continue for the next
decade. Considering the high costs and difficulties now being experi-
enced in the programming and operation of computer systems, significant
leverage can be obtained by taking advantage of each new programming
technique, even if it only doubles the effectiveness of its users.
System Software
High-Level Languages
The burden for ease of use falls on high-level languages. Such
current programming languages as COBOL and FORTRAN will be enhanced to
take advantage of new easier to use features as they appear in sub-
systems, such as communications and data base management. However, the
1980's are likely to see a shift away from procedural languages toward
those that tend to describe a problem rather than state the solution.
There now exist rudimentary problem-definition languages or dialog
processors that are used to tailor an applications package--a payroll,
for example--to the needs of a specific user. Many more generalized
products will appear through the 1980's, which will provide significant
advances in ease of use. Most systems try to be "English-like" (as,
for instance, COBOL), so ease of use and intelligence are generally
measured by how well the vocabulary and grammar of the product match
those of the user and the application. Much more success Is being
achieved in matching the application than the user.
The mid-1980's should see adequate dialog processes for computer
operation, data definition, report generation, generalized query-update,
job control language definition, and many applications areas. By late
1980's there may be some really significant English-like language
recognition, with the system resolving ambiguities through dialog with
a user and memory of that user's characteristics. It may be possible
then to contemplate the body of knowledge that would have to be encoded
to provide a system with intelligence.
It might be useful for SSA to study in detail which high-level
languages would be most appropriate for the long term. In this connec-
tion, attention is called to the study underway in the Department of
Defense looking toward the standardization of certain high-level
languages for particular uses. The SSA might consider a similar study
78
to decide on the appropriate languages for its use in applications
software and systems programming.
Operating Systems
Many of the functions now performed by operating system software
are likely by the mid-1980's to be perforated by computer microcode.
The major functions that remain, such as job scheduling, non-shareable
device allocation, error monitoring, and recovery, will be performed by
relatively simple monitors dedicated to specific modes of operation
(e.g., batch, time-sharing) in some form of virtual machine environment.
Evolution to this functional pattern will be slow, but the trend is
obvious.
System Management Software
By 1985, say, computer systems should automatically log and report
the data needed to control related external activities, including tape
and disk library control, external job scheduling, and user accounting
and billing. Logging will also be automatic for references to protected
files. The file management system will control access symbolically, and
the logging system (inaccessible to most users) will record all
references. This capability, a subset of the automatic recovery logging
process, should provide adequate file access control for multiple users.
Measurement facilities for system performance will be needed, in
addition to basic logging facilities, so that managers can observe the
performance of programs, the balancing of system resources, and so forth.
Such measurement facilities probably will interface with the diagnostic
and error-detection software. System manufacturers and specialized
software firms have developed competent performance measurement software.
Little further evolution is needed for adequacy of measurement at an
overall level. System simulation software, used to help users predict
the behavior of changed systems and configurations, will be based on the
results of the measurement software.
Software Costs
The trend toward separate pricing of software is expected to
continue. The operating system may be priced separately, but this strat-
egy is currently in the evolutionary stage among the suppliers and not
as clear as with other software components. Other varieties of software
will be separately priced. Prices will vary by the function and the
level of computer system for which they are designed. For the large
multiprocessor system of 1985, the following-software prices are
forecast (in 1977 dollars): -
Data Management System
Language Processor (each)
System Management Complex
Message Control Program
$60,000
$12,000
$60,000
$50,000
79
These are generally higher than prices for equivalent products today
because of their greater value and complexity. The data management
and system management software will often dominate a user's involvement
with the computer.
Central Processing Units
l
By 1987, say' the predominant technology used in the design of
Central Processing Units (CP0's) is likely to be Very Large Scale
Integration. This technology will be based on improvements in inte-
grated circuit design and fabrication. Within a decade, this
development will enable from 10,000 to 50,000 logic gates to be placed
on a single integrated circuit chip, with performance equivalent to
that of today's most powerful types (subnanosecond emitter-coupled
logic). VLSI should decrease the cost of today's mid-range mainframe
CPU's by one to two orders of magnitude. Upper performance limits are
not as clear, but the large decrease in gate-to-gate interconnection
distances made possible by placing such large numbers of gates on a
single chip should result in hardware performance two to four times
greater than that attainable with present technology.
Using VLSI, the physical size of an equivalent mainframe CPU will
be reduced drastically. Today's freestanding mainframe computers would
be equivalent in size a decade from now to today's desk-top micropro-
cessors. The size of the CPU's will be determined by the human
interface devices, such as the keyboard input and the cathode ray tube
display terminal, and not by the internal CPU logic.
System Architecture
Future architectures will stress direct high-level language-based
execution, protection and security of data and data independence, as
well as an increasing trend toward dedication of functions such as
input/output control and file management. A major thrust will be toward
systems organizations that can interface flexibly with formalized
communications networks and allocate program execution among various
locations. The large centralized processing complexes of today will be
replaced by very loosely coupled and highly distributed systems that
will be used in an on-line fashion, as opposed to the batch and time-
sharing modes of usage prevalent in current systems.
Processing functions that require access to large amounts of
shared data or extremely large amounts of specialized processing power
will be centralized and available through communications links.
Examples of such centralized functions are data base management systems
requiring access to large amounts of shared data and specialized arrays
of scientific processors, which are unique to a specific type of data
processing capability.
Distributed processing and parallel processing will have large
roles to play in the computer of the future. However, it would be wrong
to assume that multi-programming systems of the type in use today are
incapable of further development. Many of the problems experienced with
80
such systems arise from the extreme disparity in speed between their
high-speed memories and the fixed head disks or drums used for data
storage. Such problems will disappear when either fixed head disks or
drums are replaced by bubble or charge-coupled memories or when high-
speed semiconductor memories become large enough for the disks and drums
to be dispensed with altogether.
The difficulty in forming a clear view of the direction in which
office data processing will develop is that there are so many options
open. Office data processing is used to cover accounting, inventory
control, invoicing, and the like, and to exclude linear programming,
economic modeling, and similar applications. Even though the latter
may be pursued in an office environment, they are considered to be of a
scientific nature. Computations to be performed in office data
processing may readily be broken down into small packages. There is no
compelling need for fast processors with very large memories; many
tasks can be accomplished with distributed systems using microprocessors
with modest memories.
On the other hand, the advantages to an organization of centralizing
its data processing operations are also apparent. Office data processing
is concerned with the handling of data and at this moment the state of
data base technology is in rapid flux. This results from the development
of large capacity disk files and the emergence of a need for some advances
over the filing systems developed since the late 1950's.
Microelectronics will enable the benefits of electronic data
processing to be brought to small enterprises, including those that
employ only a few clerks. It is not hard to imagine machines resembling
the multi-register accounting machines of the pre-electronic era but
containing powerful microprocessors. In the case of a small office,
these would plug into a low-cost disk unit or bubble memory. The same
machines would have applications in larger offices, where they would be
connected to an agency-wide network. Computers and computer-based
terminals developed for the growing market are likely to become available
in quantity and at low cost. They may be expected by their very existence
to have a broad impact across the whole computer field.
SEMICONDUCTOR TECHNOLOGY
The increased complexity and improved cost/performance of semi-
conductor devices have closely paralleled the progress in cost/
performance of computer hardware over the past quarter century.
Advances in semiconductor technology will continue to make marked
contributions to the evolution of computer system in the decade ahead.
The primary impact, though, has shifted from the central processor to
the terminal, peripheral, and memory areas.
Integrated Circuits
Because of problems in electronic interconnections such as power
loss, and noise, the packing of functions into integrated circuits
(IC's) has become the industry's method of making more powerful and
81
reliable computers. The major costs involved in IC production are
(1) making the silicon chips and (2) assembling and testing the devices.
In manufacturing, because of yield losses, increasing the number of
functions--and chip area--raises the cost of the chip exponentially.
In assembly and testing, the cost is relatively independent of the
number of functions. The cost per function of IC manufacture has a
curve in the shape of a U--the sum of the two other curves--with a
minimum cost point (Figure 10, page 82~. As the manufacturing process
improves and yields get better, the minimum cost point moves to the
right. In general, the complexity of products at the minimum cost point
has doubled every year since the introduction of the integrated circuit.
This means that within 20 years, if the present rate continues, IC's
will be available with 1 billion elements.
Attaining higher levels of integration has so far been achieved
primarily in three ways:
Increasing chip size by reducing the random defects that cause
yield losses.
Introducing circuit innovations allowing higher function
densities.
Making individual circuit elements smaller.
Making circuit elements smaller has been the primary method of
increasing integration so far and will probably remain so in the near
term. Reductions by a factor of two have been occurring every five
years. As dimensions are decreased, speed and density increase, but
power density does not. This means that circuit densities could
increase by a factor of 64 and speeds by a factor of eight over the
next 15 years.
Combined density and chip size extrapolations indicate an ultimate
increase of functional complexity by a factor of 2,000 during the next
15 years, with costs increasing only slowly from those of today's
complex chips--resulting in cost per function drops of 100 (or even
1,000) to 1.
Microprocessors
-
As semiconductor technology has developed over the last 15 years,
it has become the cornerstone of the electronic industry. In the last
five years, the advent of mass-produced microprocessors has accelerated
the pace. Figure 11 shows past and future cost trends for transistor-
transistor logic (TTL) microprocessors.
The cost per active element group (AEG)--a measure equivalent to
logic gates and memory cells--of transistor-transistor logic (TTL) has
been reduced by a factor of 60 in the past ten years, while assembled
TTL has been reduced by a factor of only 15. Cost reductions for
interconnection and packaging have not kept pace, but microprocessors
have broken much of the cost barrier by decreasing the number of system
assembly operations.
82
in
o
10.00
<~' 1.00
LLI
a:
`~, 0. 10
J
o
0.01
0.001
\\\ Total Cost //
/ / Silicon Chip Cost
, ~ ~ , (Increases with
\ \ // Complexity)
\\ / /
\ _ _ ,
\,
\` /
~_
Assembly and Test
Cost (Decreases per
Function)
N NUMBER OF FUNCTIONS PER CIRCUIT
Minimum Cost Point: As
technology improves
the component cost
curves move to the
right and the minimum
cost point reaches a
higher number of func-
tions per circuit.
FIGURE 10 Integrated Circuit Production Cost
YEAR
'67 '68 '69 '71 '73 '75 '77 '79 '82 '85
I I I I I I 1 1 ~ 1 1 1 1 1 1 -'
Assembled TTL
Or Lonic GrouD
_ _ _ _
<__
Microprocessor ~ ~
—_ _ _
CUMU LATI VE AEGS (mil lions)
10~ 106 : 107
FIGURE 11 Cost Per Active Element Group For Transistor-Transistor
Logic and Microprocessors
83
The microprocessor first integrates the processor, then memory (now
called the microcomputer) and later peripheral functions.
Increases in functional capability, shown in Figure 12, result from
increasing the component density on a silicon chip by advances in:
· circuit architecture
· device structures
· processing technology
· imaging techniques
Further progress will be made in circuit architecture and device
structures, but much of the future improvement in AEG's per chip must
come from processing and imaging.
The 4-bit microprocessor appeared in 1971 and the 8-bit micro-
processor was in production two years later. The 16-bit microprocessor
was introduced in 1975, and a 16-bit microcomputer with 32,000 bit
memory is forecast by 1980. With reasonable confidence, the industry can
predict that by the 1980's it will have the technical capability to build
a single chip 32-bit microcomputer with 1 million bits of memory.
The rising curve in Figure 13 shows the increasing number o f active
element groups (AEGs) per chip as a function of time.
The functional equivalent of a medium-scale computer, depicted in
Figure 12, cost $30,000 in the early 1960' s . Its equivalent now has
dropped to $4,000 and is projected to be less than $100 by 1985,
putting it in the price range of the personal computer market. As this
is accomplished, greater challenges will be encountered in the costs of
sales, service, and maintenance, requiring that the industry learn to
incorporate self-diagnostic and self-repair functions into systems.
Figures 10, 11, and 12 show the expected cost and complexity improve-
ments of microprocessors.
· During the next decade cost per circuit element will decrease by
a factor of LOO.
Cost of a specific product, such as the medium scale computer,
will decrease by a factor of 50.
Circuit complexity of microprocessor chips will increase 100
times.
Functional complexity will increase from a 16-bit micropro-
cessor to a 32-bit microcomputer with 1 million bits of memory.
The most significant change in microprocessors will be in
the reduced number of chips or devices required in a system. The one
chip microcomputer, having both the memory and processor on a single
chip, is the leading edge of this trend. As circuit density increases 3
the amount of memory will increase, with peripheral and ~nput/outpuc
functions, such as analog to digital converters, also included on the
chip. This increase in function on the chip will reduce the cost of the
84
1011r
I~ 109
UJ
Q
an
3 107
CC
A
llJ 5
~ 10
111
J
111
'_ 1 0
0
1 960
1o6
05
104
I
103
1o2
~0
1 .0
0.]
Resolution Limits
..:::.X-Ray......
..................................
A E-Beam - .... ......
64K RAM
1 6K RAM I
, ail - ·- , .~. I . ..~...~ . -
4K RAM O ~ ~
-
-
-
1 -Chip
Calculator 0 ~
1 6-Bit
~ Microprocessor
-
32-Bit
Microcomputer.
1000K-Bit Memory
.,
16-Bit Microcomputer.
32K-Bit Memory
970
1 980
YEAR
FIGURE 12 Semiconductor Chip Complexity
rrc Ice
1K RAM
_ ,~5\'
,_ ~ _ ~
/
1~ occults I ._ Svstems-
1 990
4K~.:
~ _~5
~~ ~ ~~—Microprocessor
_
Micro-
computer
-
,_ _
-
'
1960 1965 1970
1975 1980 1985
YEAR
FIGURE 13 [Distributed Semiconductor Power
05
Cal
1 04 O
1 .... ~
10 , D
:D <
1 o2 (* Z
1Q
85
New Applications
.
Electromechanical Logic Replacement
SC Logic Replacement
-
1970 1975 1980 1985
YEAR
FIGURE 14 Microprocessor Application Evolution
end product and open new markets for microcomputers unavailable today
because of their costs.
The product and application impacts of microprocessors have been
very profound. As shown in Figure 14, the first microprocessor
applications were as replacements for semiconductor logic devices. The
data processing industry was the leading user of microprocessors. As
microprocessors have become faster and more powerful, data processing
products such as minicomputers, small business computers, terminals and
peripheral devices now use large numbers of them.
By 1975, microprocessor prices had declined enough to start replac-
ing electromechanical devices. The appliance industry is now progressing
rapidly toward microprocessor-controlled products. The transportation
industry will probably be the next major microprocessor user. By 1980,
it is likely that automobiles will use millions of microprocessors to
control engine functions. The "smart" telephone will be an important
future application; the installed U.S. base of telephones is 150 million
units, with yearly additions or replacements of about 6 million units.
Other large potential microprocessor applications are the control
of TV's, tape recorders and record players. A current entertainment
product is the programmable video game. It is likely to evolve into the
home computer. These and other potential applications are summarized in
the following table.
86
Microprocessor Application Potential
MILLIONS OF MICROPROCESSORS
l
USED ANNUALLY*
PRODUCT LOW HIGH
Data Processing Equipment 8 10
Business Equipment 3 4
Consumer Equipment
0 Appliances 20 30
· Audio-Visual Equipment 20 30
· Phones 2 6
~ Other 4
Transportation Equipment 10 15
Communication Equipment 1 2
Industrial Equipment 1 2
Miscellaneous 1 1
66
104
As microprocessors penetrate the energy-consuming equipment markets,
more advanced features will be implemented. Energy conservation features
will be particularly important. Microprocessors will be able to save
energy through accurate sensing and control of energy-consuming equipment
*Estimated for 1980.
.
