On Tue, Jul 23, 2024 at 7:11 AM Paul Koning via cctalk
<[email protected]> wrote:
> It's interesting that the designers of ARRA spoke about what they did, and
> were quite honest about their mistakes. Quite refreshing. Unfortunately
> that narrative is in Dutch: "Computers ontwerpen, toen".
> https://ir.cwi.nl/pub/13534/13534D.pdf One of these days I should finish my
> translation of that lecture.
ChatGPT 4o did a passable job it looks like.
C.S. Scholten
In the summer of 1947, I was on vacation in Almelo. Earlier that year, on the
same day as my best friend and inseparable study mate, Brain Jan Loopstra, I
had successfully passed the qualifying exams in mathematics and physics. The
mandatory brief introduction to the three major laboratories—the Physics
Laboratory, the V.d. Waals Laboratory, and the Zeeman Laboratory—was behind us,
and we were about to start our doctoral studies in experimental physics. For
two years, we would be practically working in one of the aforementioned
laboratories.
One day, I received a telegram in Almelo with approximately the following
content: "Would you like to assist in building an automatic calculating
machine?" For assurance, another sentence was added: "Mr. Loopstra has already
agreed." The sender was "The Mathematical Center," according to further
details, located in Amsterdam. I briefly considered whether my friend had
already confirmed my cooperation, but in that case, the telegram seemed
unnecessary, so I dismissed that assumption. Both scenarios were equally valid:
breaking up our long-standing cooperation (dating back to the beginning of high
school) was simply unthinkable. Furthermore, the telegram contained two
attractive points: "automatic calculating machine" and "Mathematical Center,"
both new concepts to me. I couldn’t deduce more than the name suggested. Since
the cost of a telegram exceeded my budget, I posted a postcard with my answer
and resumed my vacation activities. Those of you who have been involved in
recruiting staff will, I assume, be filled with admiration for this unique
example of recruitment tactics: no fuss about salary or working hours, not to
mention irrelevant details like pension, vacation, and sick leave. For your
reassurance, it should be mentioned that I was indeed offered a salary and
benefits, which, in our eyes, were quite generous.
I wasn't too concerned about how the new job could be combined with the
mandatory two-year laboratory work. I believed that a solution had to be found
for that. And a solution was found: the laboratory work could be replaced by
our work at the Mathematical Center.
Upon returning to Amsterdam, I found out the following: the Mathematical Center
was founded in 1946, with a goal that could roughly be inferred from its name.
One of the departments was the 'Calculation Department,' where diligent young
ladies, using hand calculators—colloquially known as 'coffee
grinders'—numerically solved, for example, differential equations (in a later
stage, so-called 'bookkeeping machines' were added to the machinery). The
problems dealt with usually came from external clients. The head of the
Calculation Department was Dr. ir. A. van Wijngaarden. Stories about automatic
calculating machines had also reached the management of the Mathematical
Center, and it was clear from the outset that such a tool—if viable—could be of
great importance, especially for the Calculation Department. However, it was
not possible to buy this equipment; those who wanted to discuss it had to build
it themselves. Consequently, it was decided to establish a separate group under
the Calculation Department, with the task of constructing an automatic
calculating machine. Given the probable nature of this group’s activities, it
was somewhat an oddity within the Mathematical Center, doomed to disappear, if
not after completing the first machine, then certainly once this kind of tool
became a normal trade object.
We were not the only group in the Netherlands involved in constructing
calculating machines. As we later discovered, Dr. W.L. v.d. Poel had already
started constructing a machine in 1946.
Our direct boss was Van Wijngaarden, and our newly formed two-man group was
temporarily housed in a room of the Physics Laboratory on Plantage
Muidergracht, where Prof. Clay was in charge. Our first significant act was the
removal of a high-voltage installation in the room, much to the dismay of Clay,
who was fond of the thing but arrived too late to prevent the disaster. Then we
thought it might be useful to equip the room with some 220V sockets, so we went
to Waterlooplein and returned with a second-hand hammer, pliers, screwdriver,
some wire, and a few wooden (it was 1947!) sockets. I remember wondering
whether we could reasonably submit the exorbitant bill corresponding to these
purchases. Nonetheless, we did.
After providing our room with voltage, we felt an unpleasant sensation that
something was expected from us, though we had no idea how to start. We decided
to consult the sparse literature. This investigation yielded two notable
articles: one about the ENIAC, a digital (decimal) computer designed for
ballistic problems, and one about a differential analyzer, a device for solving
differential equations, where the values of variables were represented by
continuously variable physical quantities, in this case, the rotation of
shafts. The first article was abominably written and incomprehensible, and as
far as we understood it, it was daunting, mentioning, for instance, 18,000
vacuum tubes, a number we were sure our employer could never afford. The second
article (by V. Bush), on the other hand, was excellently written and gave us
the idea that such a thing indeed seemed buildable.
Therefore, it had to be a differential analyzer, and a mechanical one at that.
As we now know, we were betting on the wrong horse, but first, we didn’t know
that, and second, it didn’t really matter. Initially, we were not up to either
task simply because we lacked any electronic training. We were supposed to
master electricity and atomic physics, but how a vacuum tube looked inside was
known only to radio amateurs among us, and we certainly were not. Our own
(preliminary) practicum contained, to my knowledge, no experiment in which a
vacuum tube was the object of study, and the physics practicum for medical
students (the so-called 'medical practicum'), where we had supervised for a
year as student assistants, contained exactly one such experiment. It involved
a rectifier, dated by colleagues with some training in archaeology to about the
end of the First World War. The accompanying manual prescribed turning on the
'plate voltage' only tens of seconds after the filament voltage, and the
students had to answer why this instruction was given. The answers were
sometimes very amusing. One such answer I won’t withhold from you: 'That is to
give the current a chance to go around once.'
Our first own experiment with a vacuum tube would not have been out of place in
a slapstick movie. It involved a triode, in whose anode circuit we included a
megohm resistor for safety. Safely ensconced behind a tipped-over table, we
turned on the 'experiment.' Unlike in a slapstick movie, nothing significant
happened in our case.
With the help of some textbooks, and not to forget the 'tube manuals' of some
manufacturers of these useful objects, we somewhat brushed up on our electronic
knowledge and managed to get a couple of components, which were supposed to
play a role in the differential analyzer, to a state where their function could
at least be guessed. They were a moment amplifier and a curve follower. How we
should perfect these devices so that they would work reliably and could be
produced in some numbers remained a mystery to us. The solution to this mystery
was never found. Certainly not by me, as around this time (January 1948), I was
summoned to military service, which couldn’t do without me. During the two
years and eight months of my absence (I returned to civilian life in September
1950), a drastic change took place, which I could follow thanks to frequent
contacts with Loopstra.
First, the Mathematical Center, including our group, moved to the current
building at 2nd Boerhaavestraat 49. The building looked somewhat different back
then. The entire building had consisted of two symmetrically built schools.
During the war, the building was requisitioned by the Germans and used as a
garage. In this context, the outer wall of one of the gymnasiums was
demolished. Now, one half was again in use as a school, and the other half, as
well as the attic above both halves, was assigned to the Mathematical Center.
The Germans had installed a munitions lift in the building. The lift was gone,
but the associated lift shaft was not. Fortunately, few among us had suicidal
tendencies. The frosted glass in the toilet doors (an old school!) had long
since disappeared; for the sake of decorum, curtains were hung in front of them.
Van Wijngaarden could operate for a long time over a hole in the floor next to
his desk, corresponding with a hole in the ceiling of the room below
(unoccupied). Despite his impressive cigar consumption at that time, I didn’t
notice that this gigantic ashtray ever filled up.
The number of employees in our group had meanwhile expanded somewhat; all in
all, perhaps around five.
The most significant change in the situation concerned our further plans. The
idea of a differential analyzer was abandoned as it had become clear that the
future belonged to digital computers. Upon my return, a substantial part of
such a computer, the 'A' (Automatische Relais Rekenmachine Amsterdam), had
already been realized. The main components were relays (for various logical
functions) and tubes (for the flip-flops that composed the registers). The
relays were Siemens high-speed relays (switching times in the order of a few
milliseconds), personally retrieved by Loopstra and Van Wijngaarden from an
English war surplus. They contained a single changeover contact
(break-before-make), with make and break contacts rigidly set, although
adjustable. Logically appealing were the two separate coils (with an equal
number of windings): both the inclusive and exclusive OR functions were within
reach. The relays were mounted on octal bases by us and later enclosed in a
plastic bag to prevent contact contamination.
They were a constant source of concern: switching times were unreliable
(especially when the exclusive OR was applied) and contact degradation occurred
nonetheless. Cleaning the contacts ('polishing the pins') and resetting the
switching times became a regular pastime, often involving the girls from the
Calculation Department. The setting was done on a relay tester, and during this
setting, the contacts were under considerable voltage. Although an instrument
with a wooden handle was used for setting, the curses occasionally uttered
suggested it was not entirely effective.
For the flip-flops, double triodes were used, followed by a power tube to drive
a sufficient number of relays, and a pilot lamp for visual indication of the
flip-flop state. Since the A had three registers, each 30 bits wide, there must
have been about 90 power tubes, and we noted with dismay that 90 power tubes
oscillated excellently. After some time, we knew exactly which pilot lamp
socket needed a 2-meter wire to eliminate the oscillation.
At a later stage, a drum (initially, the instructions were read from a
plugboard via step switches) functioned as memory; for input and output, a tape
reader (paper, as magnetic tape was yet to be invented) and a teleprinter were
available. A wooden kitchen table served as the control desk.
Relays and tubes might have been the main logical building blocks, but they
were certainly not the only ones. Without too much exaggeration, it can be said
that the A was a collection of what the electronic industry had to offer, a
circumstance greatly contributed to by our frequent trips to Eindhoven, from
where we often returned with some 'sample items.' On the train back, we first
reminisced about the excellent lunch we had enjoyed and then inventoried to
determine if we brought back enough to cover the travel expenses. This
examination usually turned out positive.
It should be noted that the A was mainly not clocked. Each primitive operation
was followed by an 'operation complete' signal, which in turn started the next
operation. It is somewhat amusing that nowadays such a system is sometimes
proposed again (but hopefully more reliable than what we produced) to prevent
glitch problems, a concept we were not familiar with at the time.
Needless to say, the A was so unreliable that little productive work could be
done with it. However, it was officially put into use. By mid-1952, this was
the case. His Excellency F.J. Th. Rutten, then Minister of Education, appeared
at our place and officially inaugurated the A with some ceremony. For this
purpose, we carefully chose a demonstration program with minimal risk of
failure, namely producing random numbers à la Fibonacci. We had rehearsed the
demonstration so often that we knew large parts of the output sequence by
heart, and we breathed a sigh of relief when we found that the machine produced
the correct output. In hindsight, I am surprised that this demonstration did
not earn us a reprimand from higher-ups. Imagine: you are the Minister of
Education, thoroughly briefed at the Department about the wonders of the
upcoming computing machines; you attend the official inauguration, and you are
greeted by a group explaining that, to demonstrate these wonders, the machine
will soon produce a series of random numbers. When the moment arrives, they
tell you with beaming faces that the machine works excellently. I would have
assumed that, if not with the truth, at least with me, they were having a bit
of fun. His Excellency remained friendly, a remarkable display of self-control.
The emotions stirred by this festivity were apparently too much for the A.
After the opening, as far as I recall, no reasonable amount of useful work was
ever produced. After some time, towards the end of 1952, we decided to give up
the ARRA as a hopeless case and do something else. There was another reason for
this decision. The year 1952 should be considered an excellent harvest year for
the Mathematical Center staff: in March and November of that year, Edsger
Dijkstra and Gerrit Blaauw respectively appeared on the scene. Of these two,
the latter is of particular importance for today's story and our future
narrative. Gerrit had worked on computers at Harvard, under the supervision of
Howard Aiken. He had also written a dissertation there and was willing to lend
his knowledge and insight to the Mathematical Center. We were not very
compliant boys at that time. Let me put it this way: we were aware that we did
not have a monopoly on wisdom, but we found it highly unlikely that anyone else
would know better. Therefore, the 'newcomer' was viewed with some suspicion.
Gerrit’s achievement was all the greater when he convinced us in a lecture of
the validity of what he proposed. And that was quite something: a clocked
machine, uniform building blocks consisting of various types of AND/OR gates
and corresponding amplifiers, pluggable (and thus interchangeable) units, a
neat design method based on the use of two alternating, separate series of
clock pulses, and proper documentation.
We were sold on the plan and got to work. A small difficulty had to be
overcome: what we intended to do was obviously nothing more or less than
building a new machine, and this fact encountered some political difficulties.
The solution to this problem was simple: formally, it would be a 'revision' of
the A. The new machine was thus also called A (we shall henceforth speak of A
II), but the double bottom was perfectly clear to any visitor: the frames of
the two machines were distinctly separated, with no connecting wire between
them.
For the AND/OR gates, we decided to use selenium diodes. These usually arrived
in the form of selenium rectifiers, a sort of firecrackers of varying sizes,
which we dismantled to extract the individual rectifier plates, about half the
diameter of a modern-day dime. The assembly—the selenium plates couldn't
tolerate high temperatures, so soldering was out of the question—was as
follows: holes were drilled in a thick piece of pertinax. One end of the hole
was sealed with a metal plug; into the resulting pot hole went a spring and a
selenium plate, and finally, the other end of the hole was also sealed with a
metal plug. For connecting the plugs, we thought the use of silver paint was
appropriate, and soon we were busy painting our first own circuits. Some time
later, we had plenty of reasons to curse this decision. The reliability of
these connections was poor, to put it mildly, and around this time, the
'high-frequency hammer' must have been invented: we took a small hammer with a
rubber head and rattled it along the handles of the units, like a child running
its hand along the railings of a fence. It proved an effective means to turn
intermittent interruptions into permanent ones. I won't hazard a guess as to
how many interruptions we introduced in this way. At a later stage, the
selenium diodes were replaced by germanium diodes, which were simply soldered.
The AND/OR gates were followed by a triode amplifier and a cathode follower. A
II also got a drum and a tape reader. For output, an electric typewriter was
installed, with 16 keys operable by placing magnets underneath them. The
decoding tree for these magnets provided us with the means to build an
echo-check, and Dijkstra fabricated a routine where, simultaneously with
printing a number, the same number (if all went well) was reconstructed. I
assume we thus had one of the first fully controlled print routines.
Characteristic of A II’s speed was the time for an addition: 20 ms (the time of
a drum rotation).
A II came into operation in December 1953, this time without ministerial
assistance, but it performed significantly more useful work than its
predecessor, despite the technical difficulties outlined above.
The design phase of A II marks for me the point where computer design began to
become a profession. This was greatly aided by the introduction of uniform
building blocks, describable in a multidimensional binary state space, making
the use of tools like Boolean algebra meaningful. We figured out how to provide
ARRA II with signed multiplicative addition for integers (i.e., an operation of
the form (A,S) := (M) * (±S') + (A), for all sign combinations of (A), (S), and
(M) before and of the result), despite the fact that ARRA II had only a counter
as wide as a register. As far as I can recall, this was the first time I
devoted a document to proving that the proposed solution was correct.
Undoubtedly, the proof was in a form I would not be satisfied with today, but
still... It worked as intended, and you can imagine my amusement when, years
later, I learned from a French book on computers that this problem was
considered unsolvable.
In May 1954, work began on a (slightly modified) copy of ARRA II, the FERTA
(Fokker's First Calculating Machine Type A), intended for Fokker. The FERTA was
handed over to Fokker in April 1955. This entire affair was mainly handled by
Blaauw and Dijkstra. Shortly thereafter, Blaauw left the service of the
Mathematical Center.
In June 1956, the ARMAC (Automatic Calculating Machine Mathematical Center),
successor to A II, was put into operation, several dozen times faster than its
predecessor. Design and construction took about 1½ years. Worth mentioning is
that the ARMAC first used cores, albeit on a modest scale (in total 64 words of
34 bits each, I believe). For generating the horizontal and vertical selection
currents for these cores, we used large cores. To drive these large cores,
however, they had to be equipped with a coil with a reasonable number of
windings. Extensive embroidery work didn’t seem appealing to us, so the
following solution was devised: a (fairly deep) rim was turned from transparent
plastic. Thus, we now had two rings: the rim and the core. The rim was sawed at
one place, and the flexibility of the material made it possible to interlock
the two rings. Then, the coil was applied to the rim by rotating it from the
outside using a rubber wheel. The result was a neatly wound coil. The whole
thing was then encased in Araldite. The unintended surprising effect was that,
since the refractive indices of the plastic and Araldite apparently differed
little, the plastic rim became completely invisible. The observer saw a core in
the Araldite with a beautifully regularly wound coil around it. We left many a
visitor in the dark for quite some time about how we produced these things!
The time of amateurism was coming to an end. Computers began to appear on the
market, and the fact that our group, which had now grown to several dozen
employees, did not really belong in the Mathematical Center started to become
painfully clear to us. Gradual dissolution of the group was, of course, an
option, but that meant destroying a good piece of know-how. A solution was
found when the Nillmij, which had been automating its administration for some
time using Bull punch card equipment, declared its willingness to take over our
group as the core of a new Dutch computer industry. Thus it happened. The new
company, N.V. Elektrologica, was formally established in 1956, and gradually
our group’s employees were transferred to Elektrologica, a process that was
completed with my own transfer on January 1, 1959. As the first commercial
machine, we designed a fully transistorized computer, the XI, whose prototype
performed its first calculations at the end of 1957. The speed was about ten
times that of the ARMAC.
With this, I consider the period I had to cover as concluded. When I confront
my memories with the title of this lecture, it must be said that 'designing
computers' as such hardly existed: the activities that could be labeled as such
were absorbed in the total of concerns that demanded our attention. Those who
engaged in constructing calculating machines at that time usually worked in
very small teams and performed all the necessary tasks. We decided on the
construction of racks, doors, and closures, the placement of fans (the ARMAC
consumed 10 kW!), we mounted power distribution cabinets and associated wiring,
we knew the available fuses and cross-sections of electrical cables by heart,
we soldered, we peered at oscillographs, we climbed into the machine armed with
a vacuum cleaner to clean it, and, indeed, sometimes we were also involved in
design.
We should not idealize. As you may have gathered from the above, we were
occasionally brought to the brink of despair by technical problems. Inadequate
components plagued us, as did a lack of knowledge and insight. This lack
existed not only in our group but globally the field was not yet mastered.
However, it was also a fascinating time, marked by a constant sense of 'never
before seen,' although that may not always have been literally true. It was a
time when organizing overtime, sometimes lasting all night, posed no problem.
It was a time when we knew a large portion of the participants in international
computer conferences at least by sight!
