(The Univac story to follow is a
long and detailed one, which I relate primarily for those who are
interested in the early days of computers. For those readers whose
interest is slight in this subject, I suggest that you read only the
sections that look interesting to you.)
The Eckert-Mauchly job
offer came through in November 1949, I believe, and I submitted my
resignation to Admiral Robinson as of the end of the current term
(mid-December). John wanted me as soon as possible so I reported for
work at the plant down on the Schuylkill River in Philadelphia on
December 15th. I was employee #137, at $7,500 a year. Although I
didn’t know it at the time, Eckert-Mauchly was ready to enter
bankruptcy. Remington-Rand, an office machines and equipment company,
had offered to take them over, assuming their liabilities and
acquiring their know-how. This happened less than a month after I
joined the company, and I found myself working for the Eckert-Mauchly
Division of Remington Rand. Mr. Arthur Draper was appointed as our
manager. Less than a week after I had reported for work, Grace Hopper
told me she didn’t want the responsibility of management, and
had asked John to appoint me as head of the programming department,
then consisting of four women who had worked as programmers on the
ENIAC project, and Grace. Remington-Rand refused to accept the prices
on the contracts with Nielson and Prudential, resulting in their
cancellation, but the Government contracts had to be honored. The
first computer was to go to the Census Bureau for the 1950 census,
and they wanted delivery as soon as possible. Nobody then realized
that ASAP was to be April of 1951.
My first major task was to recruit and train a programming staff.
Nobody then knew what qualifications a programmer should have, beyond
the obvious ones of ability to think logically and be careful with
details. But how do you measure such attributes? We looked in all the
obvious places, and very quickly came up with 12-15 promising
candidates, including my nephew, Morgan Huff. I enlisted all the
existing programming staff to assist in the interviews, and required
a near unanimous approval before hiring. We also needed a good
secretary, and a non-glamorous married woman named Mildred Beibel was
chosen. My experience in judging people’s capabilities was
entirely in the military, but our choices must have been good ones,
as every person hired grew into capable programmers, some to go on to
achieve higher levels of responsibility. Art Draper reviewed our
selections, and okayed everyone. As time went on we had to increase
our staff, and this meant establishing levels of experience,
responsibility, and pay. The programmers at first acted as their own
operators when running their problems on the computer. The time soon
came when we had to set up qualifications, job specs, and pay rates
for operators as well, and then hire those needed.
I well
remember the first programming assignment I gave the new group. It
was the task I had done by hand back at Fort Bliss when I was
paymaster for the 260th CA Regiment — the denominated payroll —
to determine how many bills and coins of each denomination are needed
to pay a work force, knowing the pay each was to receive. The
resulting programs revealed an amazing characteristic of the
programming profession — no two programs were alike! Every one
did the job, but with widely differing time and memory space
requirements. In fact, the fastest one took the most memory for the
instructions, and the most compact program was the slowest! This
established a principle that worked out far more often than not, the
trade-off between speed of execution and memory space required. For
many years, these two characteristics dominated computer design, as
both were dramatically improved with each new advance in hardware.
Univac I used a remarkable type of memory called an acoustic delay
line. Each character was represented by a set of six ones or zeros
(in 64 different combinations), and the string of zeros and ones was
converted into a complex sound wave which was sent along a tube of
mercury, being converted into electrical pulses at the far end,
amplified, pulse-shaped, and reintroduced as sound waves into the
mercury tube. Univac had seven big tanks of mercury tubes, with
capacity to "store" over 13,000 characters. These tanks
alone cost over $100,000, and compare with present-day’s
"chips" containing storage for over 16,000,000 characters
each at a cost of a few dollars, not to mention that the chip is more
than 1,000 times as fast!
After this first program had been
squeezed for all we could get out of it, we proceeded to look at our
customers’ problems, and began to study ways and means of
programming them. Our group became a resource group to produce
programs for all kinds of demonstrations, and eventually formed the
nucleus for the teams of programmers we "loaned" to our
commercial customers to get their applications ready for use. Nearly
every one of these first Univac programmers rose to the rank of
senior programmer, with responsibility for large customer programming
tasks. One of my projects to improve my programmers’
understanding of the Univac was to make a set of drawings, one for
each register or other major computer component, each on a
normal-sized sheet of paper and arranged so that they could be laid
out in jigsaw fashion to represent the logical elements of the entire
computer. I must have worked on this for six months or more, and I am
convinced it helped my people to understand better how the computer
functioned. Several years later I was amazed to see that the
maintenance engineering organization had taken these drawings,
reduced them in size, and made up a set of three large sheets as a
primary tool for looking for trouble in the computer.
For most
of 1950 we were at the development level. The first Univac was under
construction and test, and being the laboratory device it was, it
underwent many changes and improvements as it developed. Its
availability to us programmers was catch as catch can, as the
engineers always had first priority on its time. Many a programmer
burned midnight oil to get his program on the computer so he could
find out what was wrong with it, and eventually get it to run
correctly. The Census Bureau also had its programming staff, and they
too had to be given access to the computer. As the year progressed,
it became painfully obvious how optimistic the early predictions of
completion had been, and so there developed an air of constant
emergency, as one setback after another kept delaying the acceptance
test date.
Meanwhile, the office-machines-oriented marketing
organization of Remington-Rand were trying to figure out what their
company had acquired. At this time our division operated under the
research and development group called the Norwalk Laboratories, under
the direction of Lt. Gen. Leslie Groves, former head of the war-time
Manhattan (atomic energy) Project. General Groves came to
Philadelphia soon after acquisition, and turned out to be a rather
pompous individual, who was not at all popular with our people. His
deputy, Arthur Draper, however, was a smooth, almost suave,
individual, and most of us got along fine with him. The
Remington-Rand sales division was under Al Seares, who seemed to us
as a typical sales type, full of talk but short on facts. As I got to
know him later, I found a warm and honest person under his blustery
exterior. Remington-Rand had a wide range of office products and
equipment, from filing systems (the original Rand company) to punch
card equipment (the only competitor to IBM in this field) to electric
razors to typewriters (the Remington contribution). Mr. Rand was
president and, I believe, principal stockholder. He ran the company
from his executive offices at Rockledge, just over the State line in
Connecticut.
When I first came to Philadelphia, I lived in a
rented room in the Germantown area, where I eventually located a
rental house on Chew Street. We moved in March 1950, and lived there
almost a year. The only reminiscences I have are of the television
set, which the kids enjoyed, except for advertisements. As soon as an
ad would come on they would turn the dial to find a program. I warned
them that I would not have the set fixed if they wore out the dial.
Sure enough, it wore out in a few months, and was never fixed
thereafter. That was the last television set in our family until
1973. The other remembrance is of a litter of kittens being born in
the sock drawer of my chest of drawers, where we had allowed our cat
to sleep. Aside from having no yard, living in a row house in the
center city was not that much different from living in the suburbs.
We attended the Germantown Presbyterian Church, a very old and very
large church. There was a nursery and small children’s Sunday
School to which both Will and Mary Francis went, which operated from
9:30 to 12:30 on Sunday mornings, giving a wide variety of
instruction, recreation and rest for the small fry. I attended a
men’s Bible class of about seventy, and I remember the interest
when a request was received for a man to teach youths in Sunday
School. It was a real privilege to be called upon for such a task.
To keep my printers’ union card current, I
had transferred to the Philadelphia Inquirer, and dutifully put in my
one day a month. General Hamilton had urged me to remain active in
his New York National Guard brigade headquarters until after the
summer training period at Cape Cod, Mass., as he was to be the senior
brigadier there and would be in charge of all the troops in training.
He had me practically rewrite the Army regulations concerning
artillery practice and field exercises, so as to enhance his prestige
as commandant of troops, I suppose. As soon as the camp was over I
put in my request for resignation, by reason of removal from the
State of New York. Hostilities in Korea broke out in June, and the
102d AAA Brigade was in the first echelon of National Guard troops to
be called into active service, before my request could be acted upon.
Accordingly, on August 14th, 1950, 1 found myself once again at the
Armory in Brooklyn, this time to prepare for another extended period
of active duty and perhaps war service. I remember that my staff at
Univac had draped the chair in my office with an American flag as a
farewell gesture. I gave up my membership in the printers union at
this time. Meanwhile the Army brass, recognizing that certain
professionals were desperately needed in private industry, published
the news that electronic engineers (among other professionals) could
be released from active duty upon application of their companies or
government agencies. Remington-Rand applied for my release, but
nothing happened as we got ready to go to Fort Bliss, Texas, for
training. History was surely repeating itself! At 4:45pm on the
afternoon preceding the departure of our personnel on a train early
the following morning, a phone call came from First Army Headquarters
saying that I had been deferred, and was to report to that HQ the
following morning. I had already arranged for one of our junior
officers to drive my car to El Paso, with my foot-locker in it, and
he was several days on the way. I did get the foot-locker back some
months later, but not the car nor any money for it. For the next two
days I found myself at First Army HQ helping a master sergeant
prepare orders to call up luckless inactive reservists to active
duty. Then on the third day I got a call from Col. Robert Martin
(another fellow officer from my old National Guard regiment, with
whom I had attended the Coast Artillery School at Fort Monroe in
1937, and who now filled the place of Col. Linderer as special staff
officer for National Guard and Reserve Affairs for First Army),
asking me if I wanted to remain at my present work there in the HQ
office until orders could be prepared to release me, or would I
prefer to return to my civilian job. That choice was easy and I was
back in Philadelphia the next day. It took over a month to create the
regulations for releasing people like me, and my release came through
effective October 12th, giving me two full months of active duty pay,
much more than the $450 I had lost to the Veterans Administration
three years earlier!
Mary Charlotte did not like city living,
so we searched for a house we could buy, and finally bought a big one
in the Main Line suburb of Wayne. This house had four bedrooms on the
second floor and two with a bath on the third. It was by far the
biggest house we have ever had, and we luxuriated in the roominess of
both the house and the yard. The children had gone to public school
classes in Germantown, and were started in the local ones in Wayne.
However, Mary Charlotte was not satisfied with the quality of their
instruction, and arranged for them to attend a private, church-related
school in nearby Bryn Mawr. We attended the First
Presbyterian Church in Wayne, which was small enough that we didn’t
feel swallowed up in the crowds. I taught an adult Sunday School
class there for several years, and had my first encounter with people
we later learned were Fundamentalists. They were literalists, and
demanded that the Bible be taken literally, including the six days of
creation. These I had long taken to be allegorical, as science had
"proved" that the earth was billions of years old. One of
these people took me to task for saying that the story of Adam and
Eve was a myth, and was not satisfied by my explanation that the word
"myth" did not necessarily mean that it was untrue but only
that it was not historically verifiable.
The major event in 1951 was the acceptance test for the Univac I. This was to be a grueling test to determine the ability of the computer to give an acceptable volume of work. A 20-minute module of computation was chosen, in which input data from tape was transformed and rewritten on output tapes, which became the input for the next module. Fifty of these modules had to be successfully completed in 33-1/3 hours, giving a long-range performance of 50% time efficiency. Jim McPherson (the Census project director), Dr. Alexander of the Bureau of Standards, and I (representing the manufacturer) were constituted as the acceptance test board. To understand the significance of the test, one must realize that no satisfactory recording medium for high volume data had as yet been developed. Univac had developed metal magnetic tape units called Uniservos, which were supposed to be able to transfer 10,000 characters a second to or from the computer, (compared to the 10-characters-per-second rate of the paper tape reader-punch of Mark II). Also Univac was supposed to read from magnetic tape, compute on earlier read data, and write still earlier computed data, all at the same time. These were revolutionary features, and needed to be validated by the acceptance test. The test was conducted in April. It ran quite smoothly for the first 16 hours or so, getting through more than half of the 20-minute modules. Then tape errors began to become more frequent, and, since an error would necessitate a restart of the module, our performance began to drop alarmingly. Finally, the engineers decided to take back the machine and realign the read-write heads on the Uniservos, which they did. This cured the problem, and the test was successfully completed, with an hour or so to spare. Univac I was formally dedicated on June 14, 1951, by ceremonies at the plant attended by Mr. Rand (president of Remington- Rand), Dr. Peel (Director of the Census Bureau), Hon. Charles Sawyer (Secretary of Commerce), and other notables, including yours truly. Univac was now in business! Census used this machine not only for the 1950 Census but for much other work as well, until it was retired near the end of the decade, eventually to take its place along with Lindbergh’s Spirit of St. Louis and Apollo 11 in the Museum of Space and Science of the Smithsonian Institution in Washington.
The remainder of 1951 was busy with
demonstrations, development projects, and customer programming
assistance. The second and third Univac l’s were delivered to
the Army Air Comptroller and the Army Map Service. My programming
staff had rapidly become professionals, and were up to their ears in
programming assignments. From time to time we were authorized to have
additional staff, and I had my hands full trying to supervise all
these activities and interview candidates as well. In early January
1952, Howard Aiken (my former professor from Harvard who had been
hired as a consultant to the Atomic Energy Commission) appeared at
our plant to say that he had recommended that the Univac be tried as
a possible computer for AEC research computations. They had been
desperately seeking computing time on just about every computer that
was running, and had seven projects under way to build computers of
their own. He arranged for us to try a sample computation, and gave
me the data to begin programming. It was a formidable set of integral
equations in time and space, and was absolutely the most complex
computation I had ever seen or have seen since. We learned much later
that the equations defined the explosion of a hydrogen bomb, and
calculated its energy release!
Dr. Aiken arranged to get 14
days of time in late February from the Census Bureau, who had had
charge of the #1 Univac since its acceptance, but had left it in our
plant to avoid the delay to the 1950 census work that would be caused
by the move and installation at Suitland, Md., its home for the next
decade. I chose some of my best programmers, and we started
feverishly to work preparing a program. We were given much too little
time to do the programming, and we were into the second week of the
2-week period before the programs were even finished, let alone being
checked out. As the second week began, Dr. Richtmeyer arrived from
Los Alamos, with partial results previously obtained on the Bureau of
Standards SEAC computer to use to determine the correctness of our
results. On Wednesday, he announced that he had to return to Los
Alamos, even though we had not been able to match the very first
stage of the computation. He said he played second violin in an
orchestra there, and they were having a concert Thursday night that
he had to attend. So off he went, H-Bomb calculation or no! I was
working nearly around the clock, and by Thursday conked out. When I
felt refreshed enough to return late Friday, I was given the glad
news that Dick Petersen had found the (hopefully) last bug, and we
were matching the numbers left by Dr. Richtmeyer. All day Saturday
the Univac faithfully ground out its calculations, and then about 9pm
Sunday night, just a few hours before Census was to take the computer
back, the numbers exploded, and the calculation terminated. History
had been made, and Univac’s claim to be error-free in
computation was established. Much later Dr. Richtmeyer told me he
inadvertently ran the same set of data again on the Univac I that AEC
purchased, and got exactly the same results in every number he
checked, until he got tired of checking. My only claim to fame in
this computation was the setting up of the problem to use the
magnetic tapes as intermediate storage in an efficient manner —
an absolute must in those days of very-small-memory machines.
The
successful computation for the Atomic Energy Commission put us right
out in front for computer acquisition by the AEC. Nearly all the AEC
laboratories had an in-house computer construction program in being,
using the prototype design created by Dr. John von Neumann. They were
given interesting names: the ORACLE (Oak Ridge). the ILIAC (Illinois
Institute), the MANIAC (for Dr. Metropolis at Los Alamos), and the
JOHNIAC (for John Neumann at the Princeton Institute). But these
could not handle all the work, so a Univac I was purchased by Los
Alamos, two by the Livermore Atomic Research Laboratory (under Dr.
Edward Teller, the "father" of the H-Bomb and my former
Physics prof at George Washington Univ.) and one by the New York
University Atomic Energy Project. The latter agency offered me a job
as director of the computer lab at $15,000 a year. When General
Groves heard this he arranged for the leading engineers and myself to
be put on employment contracts, and I got a $6,000-a-year immediate
boost, with $1,000 a year raise guaranteed for the next five
years.
The Los Alamos group, under Dr. Richtmyer, arranged to
have me do some consulting for them on setting up new computations on
their Univac I, so I found myself on a train bound for Sante Fe (New
Mexico) with three of these scientists. They would grill me for an
hour or so, and then ask me to leave their compartment, while they
talked about the super-secret details of the problem. I would walk up
and down the car aisle until they called me back for another session.
We flew into Los Alamos on a small plane, and I stayed there several
days, with little to do, until they decided they didn’t need my
input any more. I could have visited the plant in Albuquerque where
the A-Bombs were being made, but turned it down. There are some
things one is better off not knowing! Later that year I got “Q”
clearance, so to be allowed in on the AEC secrets, but aside from one
visit to the Princeton Institute for a discussion of the H-Bomb, I
had no use for it.
Life
in Wayne was largely one of sleeping and eating for me, as my time
was almost completely taken by the fast-moving computer developments.
Even Saturdays were often spent at the plant, as the engineers often
left an operating computer available for our use. I did get to see
the wife and kids on most Sundays, the afternoons of which we spent
touring the Pennsylvania countryside. This picture of a group of kids
— (left to right) Charles Balch holding Lisa Huff, Will,
Lynne Caruthers, Caroline Caruthers, Steve Balch, and Mary Francis
— that must have been taken in 1950/51, but I don't
remember the occasion. While at Webb Institute, I had taken a
Prudential Life Insurance policy so as to be eligible for a
construction loan to build a house. Thinking that we could do the
same in Philadelphia, we bought a lot near King of Prussia, and had
an artesian well dug (for $3,000), as there was no city water there.
When I went to apply for the construction loan, I learned that
Prudential does not offer such loans in Pennsylvania. Before we could
figure out what to do about it, I was transferred to New York, and we
had to sell the lot. Meanwhile the city water line was extended, so
my costly well was useless. Fortunately, the appreciation of property
values got us our investment back, but with no profit.
On
Saturday, September 22, 1951, the day after Mary Charlotte’s
birthday, I was working in the yard when I noticed her disappearance.
I searched all over the yard and house for her and finally found her
on her bed, crying. I asked her, "What on earth is the matter!”
She replied, “You forgot my birthday!” I responded, “But
your birthday is tomorrow!” She said, “No, it was
yesterday.” What saved my neck was that I had bought her card,
and when I got it from my bureau drawer and showed it to her, she
realized that I had not forgotten, but had only gotten the date
wrong. Was that ever a close shave!
With the kids now in
school, Mary Charlotte took a job with Pennsylvania Bell as a long
lines operator, and continued that work until we moved to Connecticut
two years later.
Another
wedding occurred October 25, 1952, when Margery Huff became Mrs. Tom
Henney. Although we didn’t know it at the time, she was to be
the largest contributor to the Mitchell clan, with her nine children,
most of them now married with children of their own.
In August or early September of 1952, Walter
Cronkite of CBS approached Al Seares (Remington-Rand’s VP for
Sales) with the idea of having Univac predict the outcome of the
forthcoming Presidential election. Our sales people jumped at the
chance of this free publicity and persuaded Mr. Rand to provide an
adequate budget. My programming staff was dedicated to the job of
preparing the necessary programs, and Dr. Max Woodberry, professor of
statistics at the University of Pennsylvania, was hired as a
consultant to provide the prediction techniques. Woodberry had
studied all the previous national elections in US history, and had a
good grasp of the requirements of the present effort. He soon had a
goodly number of our people pounding desk calculators to compute all
sorts of criteria data by which to measure the accuracy of returns
from the polling places. There would be no time to print and
proofread these numbers on election night. Even with these
safeguards, we had some horrendous errors get through the system! We
programmers were assigned tasks by Dr. Woodberry, without really
understanding much of what we were supposed to be doing, but we
worked around the clock on our individual runs, most of which didn’t
get checked out until just before election day. In fact, the
integrated system passed its first okay checkout late that afternoon.
CBS had brought its television crew to our Philadelphia plant several
days before the Great Day. Their presence contributed to the
confusion, as the engineers wanted to groom the computer and we
wanted to test programs, while the TV engineers were trying to check
out their set-up. CBS had also set up leased telephone lines from all
over the country to pipe the data to us, with people at every polling
place to input the returns data. They had provided us with three
top-notch typists to record the data on magnetic tape with our new
Unitypers — three as we had the computer compare the three
inputs of each returns number, and accept it if any two or more
agreed. Mr. Rand had had a special typewriter prepared with very
large type, so the television camera could pick up the actual typing
of the first prediction as the computer produced it.
All was
finally in readiness about 8:30 election night, and we ran the first
few returns into the computer. Jim McGarvey, our operator, apparently
got upset by the knowledge that the entire nation was watching him on
their TV’s, and goofed up the type-out. Walter Cronkite
immediately took the camera off the computer and ad-libbed until we
got a print-out, which said that Eisenhower would win by more than
10-to-1. Since the pollsters had forecast a close election, and
Woodberry had no reason to have confidence in the correctness of our
programs, he decided that we had made an error in the determination
of his trend factor. This was a statistical device by which he could
predict the performance of States from which no returns had yet been
received. Its calculated value was typed out as 10.1. Thinking our
program must have made an error in decimal place, he had us replace
the calculated value with the value 1.01, making Eisenhower still the
leader, but just barely. This was the prediction which Cronkite gave
to the TV audience. Before we could get another prediction, we
noticed that the total vote being processed was jumping erratically,
sometimes higher, sometimes lower than before. I immediately inquired
of my programmers as to what changes in program had been made, and
found one had been made -- a small change. I had him replace his
changed program with the original one, and we once again had
substantial and believable total returns. These were made the basis
for a second prediction at about 10:30. Since the trend factor used
was still 1.01 for States with no returns, the prediction was
stronger for Eisenhower, but by no means a landslide. It was not
until after midnight, when the populous West Coast States returns
were in that the landslide was proved, as by now the trend factor had
dropped out of the computation. Fortunately for us, a reporter from
Associated Press had actually seen the early printout of the
landslide prediction, and made that the feature of his news story the
next day. For the next two years, Univac was in the news in one way
or another all over the world as the computer nobody would believe! I
had occasion several days later to type out the details of that first
prediction, in terms of total vote and electoral college vote
predicted. In all, Univac correctly predicted 44 of the 48 States
from those one million plus early returns from only eight or so
States. Of the four not correct, Tennessee was undecided for over two
weeks, New York had an erroneous return for Stevenson of several
million votes from a new precinct in the Bronx (which we could not
screen), and the other two were very close. Since Dr. Woodberry was
too busy to do so, John Mauchly wrote a paper for the American
Statistical Society about the prediction techniques used. He had me
actually write the final paper, although I can claim no credit for
the contents.