A REVOLUTIONARY TECHNOLOGY all dimensions of this trigger are less than 3.

Physical merging of storage, switching

one eighth inch by one eighth inch. and information transmission. Only the
Fig. 6 illustrates and characterizes what merger of logical and systems design with
is believed will be a revolutionary tech- Reasoning from this example, the im- microminiaturization and batch-bulk proc-
nology. Cryogenic elements have been pending revolution in computer tech- esses will produce the full impact of the
chosen for purposes of presentation, but nology will be based on: revolution.
several different types of devices under 1. Microminiaturization. Extremely
consideration by research groups would small, active, passive, and interconnecting This revolution will have far reaching
have served equally well. The small elements allowing dense packing to meet implications in all phases of the computer
squares in the illustration represent logical logical complexity and speed requirements, field, starting with research and proceed-
elements. For easy visualization all the 2. Batch-bulk processes. As character- ing through system design, manufactur-
elements shown in Fig. 6 have been magni- ized by Professor Buck, it is anticipated that ing, including the user. It is the purpose
fied. The expanded picture of one ele- systems will be manufactured by producing of this meeting to bring attention to the
interconnected batches of circuits from bulk
ment shows as an example, a trigger cir- raw materials in automatically controlled impending changes and to provoke discus-
cuit with its interconnections. The over- continuous processes. sion.

ease of coding, especially through auto-

Computer Design from the Programmer s matic programming techniques. The
importance of the computer design from
the programmer's point of view is
Viewpoint obviously great in the case of the general
purpose computer or data processor.
It is only somewhat less true in the case
W. F. BAUER of the stored program computer designed
for more specific application.
sophistication made possible by the For present purposes some of the
M R. RICE HAS given a stimulating
introduction to the subject of the
impending revolution in computer tech-
machines of more advanced design. In
the United States today, a condition of
factors of modern computer design will
be discussed and what may be considered
nology. This paper will continue this insufficient manpower to program com- to be the "ultimate" in computer design
introduction by calling attention to the puters efficiently is rapidly being ap- from the standpoint of the user will be
spectrum of activity in computer design. proached; if adequate strides of progress described.
On the one extreme of this design spec- in programming are not made, more and
trum is the user and on the other extreme more people with less qualification will
is the engineer-designer. Interpolated be brought into programming, and the Design Features
between these two positions are the costs will rise rapidly as a consequence.
There has been much discussion over
following: problem formulator, program- The author believes that the limiting
the past years on trends in instruction
mer, system specifier, system designer, factor in computers today is not the
repertoires of stored program computers.
logical designer, and circuit designer. switching times of logical circuits nor the
Here again the subject can be discussed
Mr. Rice has introduced the subject access times of memory units. Rather,
in terms of a spectrum of design pos-
from the designer end of the spectrum; the limiting factor is the lack of capa-
sibilities. On the one extreme is the
this paper will make some further bility, on the part of machine and on the
micro-instruction which is not powerful
remarks of introduction along the lines part of the user, for advanced program-
ming and advanced application. but is universal in character; it provides
of the user's viewpoint and then further
a small unit building block for the
develop ideas of what the user expects in Attempts have been made to pin down computer program. On the other extreme
the way of computer design. the idea of the productivity of the is the macro-instruction which is more
From the standpoint of the user the programmer and to determine how this powerful but also more restricted in
principal implication of the impending productivity has changed through the application.
revolution in computer technology is years since 1950. One such "program-
As instruction repertoires become more
that information processing will cost mer productivity index" would be the
"problem-oriented" they become less
less; the advances in computer fabrica- ratio of programming cost to computer
universal in character and more special
tion and in computer system design will cost. Another such index would be the
purpose. Computer instructions more
mean that a given amount of information ratio of the size of computation staffs to
problem-oriented in nature can be synthe-
processing will cost less or, alternatively, the speed of the computer. Any such
sized through automatic programming—
for a given amount of money more definition shows a very great growth in
or, commands in the user's terminology
information processing can be purchased. programmer productivity since 1950
are translated to the conventional ma-
This in turn implies a greater sophistry and, more specifically, shows a growth
chine instruction. To illustrate how the
in computer application. Since a major factor of at least five, and very possibly
more problem-oriented machine instruc-
consideration in the application of com- as much as 25, depending on the definition
puters is programming, attention is used and the assumptions accepted.
focused on that activity. Improved This increase in programmer productivity W. F. BAUER is with Ramo Wooldridge Division of
computer technology will imply less is due mainly to the computer design Thompson Ramo Wooldridge Inc., Los Angeles,
Calif.
cost for programming because of the improvements which, from the pro-
This paper is part of a panel session on "The Im-
increases in automatic programming grammer's standpoint, make possible pending Revolution in Computer Technology."

46 Bauer—Computer Design from the Programmer's Viewpoint

 COMPUTING INPUT-OUTPUT
a CONTROL INPUT OUTPUT COMPUTING
PROCESSING a CONTROL BUFFER INPUT-OUTPUT
PROCESSING

MEMORY

MEMORY

Fig. 1 . 1 1 0 3 A - 7 0 4 - 7 0 5 system schematic

Fig. 2. I B M - 7 0 9 system schematic

tion repertoire results in its more limited

use, consider the very successful FOR- by the automatic interruption of the and a one-to-one correspondence between
TRAN algebraic formula translation sys- main computing stream; employing the the symbolic names given and the specific
tem device. In nearly all International interrupt idea is like asking a staff memory cells is effected by the computer
Business Machines Corporation (IBM) member to perform a task and then hardware. For those who would argue
704 computer installations in the west, return for another assignment when the strongly that this is not desirable or
less than 30% of the problems use first task has been completed. The feasible, consider the fact that computer
FORTRAN, and very frequently the savings in computer time and program- programming by means of absolute
use is around 10%. mer time in the employment of the machine addresses is nearly extinct in
History shows that, in general, special- interrupt idea is significant. Happily, the United States today. When a usage
purpose machines are a poor investment. this technique, first realized as a standard becomes this universal, consideration
Consider, for example, the lack of success feature in the Univac Scientific 1103A, is should be given to its inclusion as a
of computers for the solution of linear being included in such new computers as hardware item, if this is at all possible.
algebraic equations. In computer de- the IBM 7070 and the STRETCH An interesting phenomenon in com-
velopment there has been an obvious computer. puter design is the increasing memory
trend to universality. Compare, for Another new area arising is that size. Although the memory size of the
example, the 16-instruction code of referred to as the associative computer large-scale computer has increased
National Bureau of Standards Eastern memory or its cousin the indirect address. roughly as the square root of the in-
Automatic Computer (SEAC) with the The associative memory is that first creased speed, it appears that the advent
very large instruction repertoires of the introduced, developed, and used by of concurrency of input-output with
Livermore Atomic Research Computer Newell, Shaw, and Simon1-2 in their computing will change this trend. It
(LARC) and (STRETCH) computers. work with a synthesized logic theory also appears that, since large transfers
However, included in this trend is the machine. Without attempting to de- between main memory and auxiliary
march toward including the more power- scribe the technical details of this tech- memory can now be handled so efficiently
ful instruction, the instruction which is nique, let it simply be said that it allows and economically, there is a limit to,
more problem-oriented. To emphasize the programmer to set up strings of or an optimum for, the size of a high-
this, one can recall that in 1951 people quantities within the computer memory, speed memory. It appears that computer
who wanted built-in floating point were thereby allowing him to make better memories in the future will grow with
somewhat hard to find. At that time, use of the computer memory and relieving increasing computer speeds but at a rate
many users, and probably more designers, him of the necessity to specifically outline slower than the square root rate just
thought that floating-point arithmetic computer memory requirements on an mentioned.
instructions should be programmed rather a priori basis. The indirect address One of the requirements for larger
than built in. technique is developed to a considerable high-speed memory sizes lies in auto-
The real solution to the problem of extent on the IBM 709 computer and matic programming. In the system for
design of instruction repertoires for almost to an ultimate degree on the the IBM 709, for example, which allows
computers seems to be incisiveness with Gamma 60 computer. 3 This technique automatic operation of the computer
universality, the set of micro-instructions essentially allows the programmer the between and during runs on the computer,
heavily sprinkled with problem-oriented same advantages of the associative a computer control program of 4,000
instructions. memory and, in particular, allows him words in length is kept in the magnetic
The past few years have seen many to deal with names of quantities rather core at all times. It remains there for
important innovations in computer de- than the quantities themselves. the use of the programmer for the
sign. Some of these are as follows: An idea which may appear important handling of untoward situations and for
the B-box, automatic overflow and under- in computer technology from the user's performing the transition from one
flow handling, the real-time clock avail- standpoint is the symbolic memory. computer run to the next.
able for program interrogation, and the With this memory, the memory cell One of the principal faults of the
automatic interruption of the computer itself contains in symbolic form its modern computer is that it has charac-
by asynchronous devices. The last "address" or, in better terminology, its teristics of an arithmetic manipulator
innovation mentioned, "the interrupt "name." The programmer, instead of and does not have sufficient character
feature," bears extra mention and em- referring to a specific address, refers to a along the lines of information handling.
phasis. The complexity of system design symbolic name in order to summon a The modern large-scale computer is
implies that events asynchronous to the quantity from the memory. The as- used as much as 40% of the time in
computer operation (e.g., card reading, signment of names to specific hardware program debugging. In this activity the
memory interrogation, input from ex- memory cells is performed by the ma- computer is acting almost completely
ternally timed sources) must be handled chine at the beginning of the computation. as an information handling device—

Bauer—Computer Design from the Programmer's Viewpoint 47

 MAGNETIC
DRUMS

COMPUTING COMPUTING MAGNETIC

UNIT UNIT PROCESSOR TAPES

CHARACTRON
at it
T
J COMPUTING J J DATA J INPUT-OUTPUT J INPUT- J
PRINTER
CORE CORE CORE CORE
2
MEMORY MEMORY MEMORY MEMORY

rP -!-^ W ¥
CARD
REPRODUCER
MEMORY MEMORY MEMORY MEMORY MEMORY MEMORY
D E
MEMORY MEMORY

Fig. 3. Univac L A R C system

Fig. 4. "Ultradatic" system schematic

as a data processor. If one adds to this

percentage the percentage of time the characterized by such computers as computer will have important changes in
computer performs production operation SEAC, National Bureau of Standards computer programming. In the parallel
on problems which are essentially of an Western Automatic Computer (SWAC), machine, in the LARC or the Gamma 60,
information handling or data processing and Whirlwind. These computers were for example, where large components of
character, one finds that the so-called strictly sequential in their operation. the machine can be time-shared by
"scientific computer" in a scientific Input and output interrupted the main problems, it may be possible to use the
environment is performing data process- computation. Frequently, a megacycle machine to perform automatic program-
ing and information handling upwards bit rate in the main computation was ming processing as a fill-in operation
of 70% of the time. Yet neither instruc- replaced by a 25 bit-per-second rate for during the normal course of the work of
tion repertoire nor system design factors input and output. the computer. In this case, the use of
seem to reflect this figure. A striking The second system design phase could large amounts of machine time will not
example is this: ,A great percentage of be referred to as the "buffered computer" inhibit computer automatic program-
time the information processing per- phase, characterized by the Univac, the ming developments. On the other side
formed is that of translating alphanu- IBM 701, IBM 704, IBM 705, and the of the coin is the fact that exploitation
meric input to numerical input, that is, Univac 1103 computer. With these of these machines with parallel operation
the mapping of the six-bit characters to computers, input-output required only by automatic programming techniques
series of, say, octal digits. Yet almost the time necessary for the transfer of will be most difficult. (Witness the
none of the modern computers allow the information at electronic speeds. With difficulty of the B-boxes in the develop-
programmer ease and facility in the the advent of this phase, input-output ment of FORTRAN for the IBM 704.)
handling of six-bit characters; the was taken out of the domain of me- For a thorough introduction to the
translation must always be performed chanical speeds and placed in the domain subject of programming of these new
in an indirect way, through synthesis, of electronic speeds. It should be noted style computers, the reader is referred
with extract and shift instructions. The that within the computers of this phase to a recent paper by S. Gill5.
author believes it can be safely stated there is a wide range of buffering. Fig. 1 A failure of the system design of the
that no computer in existence today shows the simplified block diagram of present computers lies in the fact that
has a system design which reflects the this computer. the duty cycle of the various components
fact that automatic programming is an
The third phase of system design can is not sufficiently high. With most
important and ubiquitous labor saving
be referred to as the "automatic com- computers, for example, the circuits for
device.
puter" represented by the IBM 709, high-speed multiplication are utilized
the TX-2 and the Univac Scientific 1105. only a small fraction of the time. In
System Design Here is seen the first concurrent operation a sense the analog computer system
of simple parallelism. With these com- design comes far closer to the ideal.
System design is the design of the puters, input-output proceeds independ- In the analog computer installation,
computer with respect to the inter- ent of the central processing. Fig. 2 for example, the computer can be spread
relation of its respective large components shows the simplified schematic of the out and "sized" to the problem. If
such as memory, input-output, control IBM 709 computer system and, inci- the average analog computer problem
unit, etc. With the advent of the dentally, shows how the memory is time- utilizes only a small fraction of the total
IBM 709 and the Univac 1105 computers, shared by the central control unit and number of amplifiers and nonlinear units,
and, more specifically, such computers the input-output control unit. a high duty cycle can be maintained on
as the LARC, STRETCH, the Lincoln The phase which is now being entered all of the equipment. Taking the lead,
TX-2,4 and the Gamma 60, there is a could be referred to as the "parallel then, from the analog computer design,
significant revolution in system design system" computer phase. This phase is it appears that concurrent operations of
taking place. This revolution certainly represented by the LARC and the large parts of the computer should be
ranks in importance with the revolution STRETCH computers just appearing. possible, and that in some way the job
taking place in the design and fabrication With these computers, there is con- for the computer must be made small
of logical elements. current operation of large components with respect to the computer. This
Consider a historical view of design. of the computer. Fig. 3 shows the idea is exploited further in the design of
Four phases of system design can be block diagram of the Univac-LARC the conjectural computer "Ultradatic"
distinguished: The first of these can computer. described later.
be called the "laboratory computer," The advent of the "parallel system" One of the goals in computer design is

48 Bauer—Computer Design from the Programmer's Viewpoint

to make the computer act in some sense installed on their own premises and Each processing unit is a small com-
more like a human. The reader has would buy time on the computer much puter in its own right. Each has registers
probably heard of "self-organizing" the same way that the average house- for variables and parameters, each has a
systems and will undoubtedly hear more hold buys power and water from utility limited number of instructions which it
of these in the future. As part of the companies. In fact, the charges for the can perform peculiar to its particular
self-organizing aspect, the psychologist services would depend on the type of function, and each processing unit re-
or physiologist could refer to the "degree services rendered. If the problem re- cords the time spent on the various prob-
of introspection" of a complicated system quired extensive use of the higher priced lems with which the computer is dealing.
in judging how "nearly human" the circuits, for example, high-speed multi- A problem number and a priority num-
system is. Certainly the human stops plying, the rental rate for the time used ber accompany each command sent from
his main stream of activities frequently would be higher than in other cases. the control units to the data processing
to reflect on what he is doing from the The user would be charged for that units. The problem number is essentially
over-all point of view. This introspection time used and only that time used, and an account number used for account-
increase can be seen in computer design. the accounting procedure would reflect ing purposes and the priority number
In a sense, the interrupt feature men- the customers' detailed use. describes the order in which commands
tioned, provides a degree of introspection Fig. 4 shows the schematic of the are to be performed by the processing
since one part of the computer system Ultradatic system. On the figure there unit in case waiting lines develop. Be-
automatically informs another part when can be seen two levels of control, the cause branch A of a problem may be com-
it has performed or when it is ready to supervisory control and the normal pleted before branch B and the results of
perform the asynchronous operation. control unit, and a number of processing branch B are required to continue with
The large-scale computer of the future, units. There is a fast arithmetic unit to branch A, a scheme based on the sequence
by the same token, will periodically perform ultra high-speed arithmetic, and of operations within a particular problem
examine its own operations and auto- there is a computing unit to provide the would be required for interruption or de-
matically take certain steps to change lower speed arithmetic with, however, lay of operations.
its function. Ultimately it will change greater possibility for flexibility. There At first thought one might conclude
its very nature of character and thereby is a logical operations unit which special- that the scheduling problem on Ultradatic
take on a self-organizing aspect. izes in activities such as sorting and would be an impossible one; that is, it
collating. The data-translation proces- would be impossible to schedule the var-
sing unit specializes in translation and ious components so that a problem can be
"Ultradatic," A Conjectural
editing, and the input-output scheduling done in a short length of time or in a rea-
Computer
unit performs detailed control over sonable length of time. It is believed that
A computer system will be described, input and output. scheduling a problem would not arise with
which is in some sense "ultimate." The reader will notice great similarity such a computer. On the basis of the
This is done in the hope that it will in Ultradatic to the Gamma 60 computer customer's desires a certain initial prior-
stimulate thinking and comment rather which was first generally known in this ity number would be assigned to a prob-
than with the thought of foretelling the country in the spring of 1958. (The lem, and a rough estimate would be
future. The idea for "Ultradatic" Ultradatic idea was first described in made by the computer itself with regard
was motivated by the thinking described, talks given by the author to the San to its workload and as to how much time
concerning the advantages of the analog Diego and Rio Grande chapters of the would be required to perform the problem
computer system which can be sized to Association for Computing Machinery in the problem multiplexed environment.
fit the problem, and by the related idea in late 1957 and early 1958.) As the time grew nearer for the comple-
that it may be possible to make computa- In this large computer system the tion of the problem, the computer would
tion more economical by making prob- supervisory control units schedule the upgrade the priority if it appeared un-
lems small relative to the computer. multiple problems, monitor the usage likely that under the current operation
It is fostered by the idea that the large of the control units, and initiate the the schedule would be met. Thus, the
computer is more economical to use work on the problems by the control computer by means of the supervisory
today than the small computer, or a units. Each problem has a priority control would monitor itself automati-
number of small computers, as long as connected with it, and the supervisory cally and would automatically change the
the total workload is sufficient to keep control monitors and changes the pri- priorities of problems.
the large computer busy, or as long as orities according to the speed with which A very important characteristic of the
unnecessary expense is not incurred by problems are being performed. In Ultradatic computer is that it is modular
idle time. The idea further stems from more advanced configurations the super- in construction, and large components
the fact that with increasing frequency visory control would perform some kind can be added or subtracted to make up a
one sees computers being operated of introspection and, perhaps, would configuration which is currently in de-
from remote locations by means of effect the change of character of some mand at the particular installation. The
communication over voice channels, or of the units to other kinds of units processing units, shown in multiples of
at least extensive input or output of depending on the requirements of the three in Fig. 4, would be available in any
data is being fed to computers from over-all problem situation. combination and in any practical num-
remote locations. The control units interpret macro- bers. If, for example, the records showed
The central idea here is that each instructions, effect the transfer of sub- that the fast arithmetic unit had a con-
large metropolitan area would have routine variables and parameters to and tinual backlog, a new fast arithmetic unit
one or more of these super computers. from the processing units, distribute would be added to the system configu-
The computers would handle a number work among like processing units, and ration. Or, if one of the logical opera-
of problems concurrently. Organizations initiate the action of the processing tions units was found to be idle a good
would have input-output equipment units. fraction of the time, one of these units

Bauer—Computer Design from the Programmer's Viewpoint 49

could be removed from the system. T h u s pose. However, almost every installa- to modify its own program or logical struc
t h e computer can grow a n d change to tion with a small computer t o d a y has a t ture.
Concerning the comprehensiveness of in-
meet t h e changing requirements of t h e in- least one problem which should be placed structions, there is no upper bound as far as
stallation a n d the high d u t y cycle will be on a higher speed machine. Usually t h e the programmer is concerned as long as the
m a i n t a i n e d a t all times. big problem is r u n on t h e small machine instructions are grouped and presented in
inefficiently and uneconomically because such a way so that subsets of them can be
T h e m e m o r y units of Ultradatic war- used. However, there is an upper bound
r a n t certain attention and discussion. of the inconveniences of renting large
from the viewpoint of the computer seller or
This u l t i m a t e computer would h a v e a scale computer time. the computer buyer. As the comprehen-
symbolic m e m o r y ; t h a t is, t h e symbolic W i t h respect to special-purpose com- siveness goes up, the costs, of course, go up
puters, most of t h e ideas of Ultradatic are also, and the point of diminishing returns is
address of t h e cell assigned arbitrarily by
soon met. Tomorrow's technology will
t h e programmer would be stored within still applicable if, of course, attention is allow comprehensiveness through or by
t h e m e m o r y cell, t h u s allowing the pro- restricted to those of stored program de- means of self-organization.
grammer always to refer to t h a t cell in sign. Consider t h e idea of growth or of
symbolic address notation. Because of the inevitable changing character of the Jerry Mendelson (National Cash Register
t h e multiple branches of t h e programs application. A computer truly modular Company): Ten years ago we built ma-
chines with 10-microsecond adders which we
proceeding a t times a n d a t speeds unpre- in system design can be increased in power
embedded in control configurations which
dictable, a different technique m u s t to handle the control job for the new reduced the potential addition (computation
evolve for t h e use of the memory in this polymerization unit or the new cracking in general) rate from 100,000 per second to
computer. W i t h present computers in tower. Or, as a new formulation of the 10,000 per second. We then provided the
control process evolves, the computer can control system with a command list in
rocket trajectory calculations, for exam-
which about one command in ten did useful
ple, only one computer cell is reserved for be changed to efficiently handle the new work, the other nine being required to get
the height of t h e rocket. I n this com- computations. Also, ideas of system de- ready for this work. Result: 1,000 useful
puter system a cell would be reserved for sign to m a k e programming easier will be- commands per second.
t h e height of the rocket a t each quantized come more i m p o r t a n t with special-pur- For 10 years we have been smug and com-
level of height. This would be necessary pose machines. W h e n real-time control placent about having produced man's most
inefficient machine, ( 1 % efficiency), and all
since t h e height a t 9,000 feet might be re- computers appear in abundance certainly
we have done is shove this inefficient design
quired in one branch of the computations one of the limiting factors in application up the frequency spectrum in order to obtain
while the previously computed height, will be the preparation of computer pro- faster effective output, 1 microsecond adders
8,500 feet, was being used simultaneously grams. replacing 10 microsecond adders. We
appear to be heading into the millimicro-
in a completely different branch of the Although Ultradatic will probably second domain with no great effort being
computations. T h e symbolic address never appear a t the computer m a r k e t applied to changing the internal design
notation, now considered p a r t of the ad- place, some of its design ideas m a y bear characteristics which drag us down by a
dress hardware, would include indices or fruit. If not, service to computer tech- factor of 100. It's about time we paid some
subscripts to differentiate the various attention to this area, is it not?
nology advancement is still provided in
values of a variable. When the pro- unerringly pointing t h e direction n o t to Dr. Bauer: It certainly is time. I t seems
grammer realizes t h a t the height a t 8,500 follow. that too much time has been given to the
feet is no longer necessary anywhere in more local problem of circuit speed in com-
the computations, he returns the cell to parison to the time given to the global
References problem of the effect of an executed instruc-
the available unused memory pool.
tion or an executed set of instructions on
There the memory cell can be assigned to 1. EMPIRICAL EXPLORATIONS OF THE LOGIC
solving the problem to which the computer
THEORY MACHINE: A CASE STUDY IN HEURISTIC,
another problem or to another value in A. Newell, J. C. Shaw, H. A. Simon. Proceedings, is being applied.
t h e same problem. Here the associative Western Joint Computer Conference, Feb. 1957,
pp. 218-39. Robert Norman (Sperry Gyroscope Com-
memory idea would come into play; a
string of unused memory cells would be
2. THE LOGIC THEORY MACHINE, A. Newell, pany): A centralized computer facility
H. A. Simon. Transactions, Professional Group on such as you envisage, with problem solutions
established, with each cell of the unused Information Theory, Institute of Radio Engineers,
New York, N. Y., vol. IT-2, no. 3, Sept. 1956, pp. available effectively at people's telephones,
memory storing the address of still an- 61-79. runs right into a problem in human rela-
other unused memory cell, so t h a t the 3. FRANCE'S GAMMA 60, P. Dreyfus. Datama- tions.
string could be added to or subtracted tion, May-June 1958. Much business, personal and corporate,
from with convenience as required. 4. T H E LINCOLN TX-2 COMPUTER DEVELOPMENT, involves privacy. The availability of large
W. A. Clark. Proceedings, Western Joint Com- volumes of data on the private and corporate
puter Conference, Feb. 1957. affairs of the inhabitants of a community
T h e r e are certain everyday problems 5. PARALLEL PROGRAMMING, S. Gill. The Com- constitutes a threat to their privacy. It
which would be performed by Ultradatic. puter Journal, British Computer Society, 1957. seems this would result in a degeneration of
For example, once or twice per day, the 6. MODERN LARGE-SCALE COMPUTER SYSTEM the centralized computer facility to one only
DESIGN, W. F. Bauer. Computers and Automation, used for the solution of an unimportant
computer would examine all of the pro- Jan. 1957.
portion of the total volume of computation
grams in its backlog a n d would m a k e
in the community.
estimates of a gross character as to the
length of running time. Once or twice
Discussion
Dr. Bauer: This would undoubtedly be a
per day the computer would carry out a I. O. Naughton (Westgate Laboratory): problem with a system like Ultradatic.
detailed accounting and compute bills for What do you see as an upper bound (a) to However, difficulties are not insurmount-
its m u l t i t u d e of users. Of course, the the ability of a computer to modify its able. For example, the computer itself
program based on partial results? (b) to would monitor in detail all communications
computer would spend a considerable between it and a customer to insure that no
a m o u n t of its time in compiling a n d the comprehensiveness of the instructions
information was given to unauthorized
that can be provided for a programmer?
translating. destinations. Furthermore, proprietary in-
Dr. Bauer: The upper bound is the limit formation could be secured by encoding
Existence of computers like Ultradatic techniques similar to those used in military
of humans communicating with a device to
would n o t obviate the necessity for small tell it how to modify its program or logical communications today. Even today we
computers, either general or special pur- structure or to tell it how to tell itself how tend to use the telephone as if it were a

50 Bauer—Computer Design from the Programmer's Viewpoint

completely secured and private instrument, amount of programming that is done in the economical, but nobody is worried about
and it seems that our "shared" computer of application, assign some meaningful cost to how much money is being spent on pro-
the future would be at least that private. it, and assign some meaningful effect to the gramming.
amount of programming that is done. The "programmer productivity" will
E. L. Harder (Westinghouse Electric As a matter of fact I believe t h a t the probably again increase by many factors
Corporation): What is the present unit emphasis on computer design from the during the next 8 years. Weighing against
cost per computation (you define it), and standpoint of speed, size, cost, etc., and the this is the higher salaries which are being
what reduction do you anticipate as a relative lack of emphasis on computer paid to programmers in reflection of their
result of this revolution? design from the programmer's standpoint contribution to the general technology.
stems exactly from the fact that it is easy Because of these facts I would guess t h a t
Dr. Bauer: This is a very tough question to put a dollar amount on the computer the cost of computation, all things being
to answer. Everybody can define the cost costs but very difficult to put dollar amounts considered, certainly will decrease but at a
per operation of the digital computer itself. on programming costs. Much attention is slower rate during the 8 years than it has
However, it is very difficult to define the being focused on making the computer during the past 8 years.

t a k e n only after it is well s u b s t a n t i a t e d

New Logical and Systems Concepts t h a t there will be more t h a n an offsetting
gain to be obtained through factors such
as an increase in computing speed or a
R. K. RICHARDS decrease in programming effort.

B E F O R E making any a t t e m p t to
outline t h e course of future develop-
m e n t in computer systems technology,
duction of the stored program were
achieved only because there was a simul-
taneous development of relatively low-
The Course for the Future

Other a u t h o r s have already discussed

it is well to review briefly the present cost large-capacity digital storage com- the possibility of obtaining low cost logical
state of the art and how it arrived at t h a t ponents. Magnetic drums, mercury de- elements as well as low-cost storage ele-
state. T h e system designs of the first lay lines, a n d various forms of elec- ments through t h e use of film techniques.
digital computers such as t h e H a r v a r d trostatic storage were among t h e first I n view of t h e considerable advances in
M a r k I, the Electronic Numerical I n t e - storage media to be used for this purpose computer systems design t h a t were
grator and Calculator (Eniac), a n d others although magnetic cores now play t h e m a d e possible through t h e introduc-
were largely influenced b y ideas presented leading storage role. tion of low cost storage elements, it
100 years earlier b y Babbage. I n gen- N o similar progress in reducing the might be expected t h a t corresponding
eral, the system was comprised of a set cost of logical components was rea- advances might be m a d e through t h e
of decimal registers, each capable of lized however. Even today, the cost introduction of low cost logical elements.
transmitting numbers in parallel fashion of an elementary " a n d " function, for I t is the purpose of this presentation t o
to and from other registers. When a example, is generally considered to be examine the factors involved in an effort
register received a n u m b e r from another m a n y times as costly as a bit of storage to determine w h a t these advances m i g h t
source it was capable of adding t h a t in a large-capacity storage array. As be.
number to any number already contained a result of this situation, t h e p a t h of I t has been observed t h a t whenever a n
in t h a t register. Computations were systems development has been in t h e improvement is m a d e t h a t might b e
achieved b y transmitting numbers back direction of using t h e m i n i m u m possible described as being of revolutionary pro-
a n d forth among the various registers n u m b e r of arithmetic a n d control ele- portions, it has been as a result of solving
with, of course, suitable refinements to ments. T h e result has been t h e ap- a problem in one field through the applica-
obtain t h e desired results. pearance of t h e one-operation-at-time tion of a body of knowledge from another
Although the concept of long sequences general-purpose computer with t h a t one field. For example, no great improve-
of arithmetic operations with storage of operation being as simple as tolerable to ments in illumination came as a result
intermediate results was an integral the user of the computer. T h e binary of studying flames, tallows, or other
part of t h e earlier systems design, t h e n u m b e r system appeared a t this point in aspects of candles. Instead, a knowledge
stored program concept as it is known computer development. Originally t h e of electricity was used in the develop-
today was absent. Further, it is quite sole purpose (and still the major purpose) m e n t of the electric l a m p to replace t h e
likely t h a t even if t h e stored program of binary numbers was to reduce the n u m - candle. Similarly, in the field of t r a n s -
concept h a d been developed at the time ber of logical components required to portation only limited gains were realized
of construction of t h e first computers, m a k e a computer. T h e considerable by improving t h e care and feeding of
it would not have been used. T h e reason burden placed on the programmer as a horses or b y developing better breeds
is t h a t the registers used for storing n u m - result of binary numbers was considered of horses. Instead, t h e revolutionary
bers were (and still are) bulky a n d ex- of secondary importance. advances came from a study of mechanical
pensive things, either in the electro- With t h e passage of time t h e decimal engines.
mechanical or electronic versions. I t system has been appearing in a larger I n the case of electronic digital com-
did not seem feasible to think in terms of percentage of new computer designs, a n d puters, if anyone were to propose a n
more t h a n about 100 such registers in a t h e operations capable of execution b y ultimately successful concept t h a t r e -
machine, a n d this a m o u n t of storage is the newer computers are more complex
not sufficient to utilize the stored pro- a n d powerful. However, each step of
gram concept in an effective manner. progress in this direction causes an agoniz- R. K. RICHARDS is a Consulting Engineer at
Wappinger Falls, N. Y.
T h e advances in computer systems ing increase in the cost of t h e control
This paper is part of a panel session on "The Im-
development occasioned b y the intro- portion of the computer a n d is, therefore, pending Revolution in Computer Technology."

Richards—New Logical and Systems Concepts 51