UNIVAC - Software in the Early Sixties

After being hired as Director of Systems Programming at UNIVAC, I decided to begin my tenure with a tour to the four programming sites under my jurisdiction:

1. The New York operation, largely support and documentation, although I added some research capability later, in the form of Al Paster, Bill Burge. and Peter Landin.
2. The Blue Bell (PA) site, where Grace Hopper had held sway for so long (she was still there), and where the UNIVAC III was coming out as the business computer,
3. St. Paul (MN), where the former ERA people were producing the 1107 for the scientific market,
4. El Segundo (CA), where the Numerical Control people operated under Dr. Gaston Chingari.

• I knew that the former occupiers of my position had lasted 6 months on average, but they should not count on it in my case.

• I was a programmer like they were, so they should not count on fooling me as they had fooled former managers.

The Blue Bell site (outside Philadelphia) had pretty much female management, which was fine by me (I had made a precedent at Marquardt by paying women the same as men for the same work). But they were juggling the attendance records for each other like a matriarchy, which it was not. I asked for the back records, substantiated the charges, and sacked the top manager on the spot. This got their attention like nothing else might have. I appointed a David Meredith to take over.

Somehow I fortuitously inherited as an assistant one A. R. (Dick) Shriver, who understood the heavy politics of UNIVAC, and kept me from traps. And traps there were in plenty. I had never before experienced such company-disloyal infighting, and so was too naive to see the subsurface machinations. From Dick I got continuous and good advice. He later went to ICL in England, but when there were problems there he returned to the US without any money. I gave him an advance, mostly to redeem his laundry. and then rehired him. Later we rejoined at GE in Phoenix, and later still he was managing editor for (my) Honeywell Computer Journal.

My first major problem was on the software for two new computers -- the "commercial" UNIVAC III, out of Blue Bell, and the "scientific" 1107, out of St. Paul. As both Blue Bell and St. Paul software were described to me, neither seemed to afford a striking advantage against IBM, the main competitor. There was an alternate way to go, having to do with CSC (Computer Sciences Corporation). It deserves a story of its own on this site. And there is one.

An Attempted Coup

When I started with UNIVAC, we had not yet moved into the new Sperry Rand building uptown on the so-called Avenue of the Americas, but inhabited a very old building on Lexington south of Grand Central Station. Each day I was afraid that the wrought iron elevator cage was going to fall into the basement. It was there, early on in the CSC effort, that C. L. McCarty (who had inspired the CSC link) conceived a scheme to serially:

• Form his own company, taking key programmers from UNIVAC.
• Convert the CSC contract to his own, and then
• Sell that software to UNIVAC at a considerable markup.

Bob Hengen saved me another time. Jacques Stern was visiting UNIVAC. Yes, he who later became in charge of the Compagnie Bull, in France. We rode to lunch in a taxi -- Stern, his assistant, Bob Hengen, and myself. While in the taxi, Stern talked to his aide in French, and to me in English, quite unaware that Hengen was fluent in French even though I was not. And what he told the aide was how gullible I was being about his proposal. M. Stern must have been quite surprised when I declined his proposition.

Hardware Problems

Because software is how the user sees the computer, I had to be sure of the hardware as well. IBM had a Service Bureau where we could buy some time. I had someone write a program to exercise the full range of floating point numbers. We ran it in this equipment, and then we ran it on the 1107. The IBM machine worked from 10 power -38 up to 10 power +38.

The 1107 worked up to 10 power +38 at the high end, but it mushed out at 10 power -22 at the low end. This was unacceptable. A competitor had to be able to get the same answers to problems that IBM did! So I called St. Paul Engineering, finding that this was due to underflow in the more significant part of the number. I then issued an edict that it had to be changed to underflow only on the less significant part.

Then came a strange and complaining phone call from the wife of Tom somebody who had done the design. She insisted that her husband had done it right. I allowed that perhaps he had, but I still thought that we needed to get the same answers that IBM did.

The 1050 Computer

After joining UNIVAC, the first new machine to come along was the 1050, a quite small system, for which Eric Clamons (later designer of the TEX language) was the product planner. Due to our former cooperative work on code standards, the 1050 was the first ASCII-based machine, one of the few until PCs arrived.

The building of the PAL assembler for the 1050 was given to a new man (I cannot remember his name) in Blue Bell. Several of us gave advice, and when it arrived for use it was great. I called the man and told him he was about to get a 25% raise for doing such good work. His boss's boss, David Meredith, who reported to me, said to me that he would not process such a raise, as the man had been with UNIVAC for only 3 months, and could not be considered for review until at least 6 months! I said he would process it, or I'd be finding somebody new for his job that would! He did. That's how we got good software in the early days, folks! The man called to thank me, and said "I'll work even harder now".

Many people were pleased by that assembler. I think it increased sales greatly. Paul Nowak of Honeywell's Federal Systems Division sent me this E-mail in 1993:

"Bob, my first language was PAL on Univac 1005/1050. It was actually possible to execute a RASD - Read And Score Drum ... I dropped the heads on a FASTRAN drum by feeding drum commands to a drum macro labeled with the name of the tape macro."

Punch Cards

It is possible to make decisions about obsolescence in the computer business. During my tenure at UNIVAC, I went to Pres Eckert (he of the ENIAC, then a VP there) and said that we just could not afford to keep on making and selling cards with round holes in them when IBM was using cards with rectangular holes. He agreed, and we passed our decision on to the President, Dr. Louis Rader. UNIVAC cards used rectangular holes from then on. We wiped out an entire product line! It was a lot easier than accomplishing the same via an X3 standards committee.

This may seem a trivial, if somewhat expensive, action to take, but it had a profound effect later. When Eric Clamons was running the ASCII committee in X3 he found almost insurmountable opposition from IBM. He solved it by linking ASCII and EBCDIC to each other via the punch card hole representations, quite independently of their differing binary encodings. This one-for-one relationship allowed IBM a certain amount of face and independence.

And -- because every ASCII character was linked uniquely to an EBCDIC character via a standard punch card code that each now had in common, IBM was able to build a chip that translated in either direction between ASCII and EBCDIC. Today, friends, this is the basis of your being able to use ASCII PC terminals connected to EBCDIC mainframes! Aren't you glad I took that initiative?

The 418 Computer

One Lee Johnson was the Vice President for Federal Systems for UNIVAC, and therefore headquartered in Washington, DC. It was said of him that his usual method for getting agreement to his proposals was to provide plenty of coffee and then lock the doors to the meeting room. His was a naturally overbearing personality; most of us ascribed it to what 1990s people call "being vertically challenged".

It happened that the St. Paul engineers had come up with a design called the 418 computer. They proposed in a meeting that my groups do the software for it. My argument against that was that we had not enough manpower to add that to the workload we already had, which would suffer. Moreover, I estimated the software costs at $3 million. At this point, Lee said that, to the contrary, it was not a large task, and he had programmers that were virtually standing around in Washington with not much to do. They could polish off the 418 software suite in short order.

Buying Lee's argument that the software could be done for a song, Dr. Rader succumbed and signed off. The next day the Washington newspapers were full of ads for programmers! After I left, Milt Bryce had carried on with our software production control, and sent me a sample of the printouts. Perhaps inadvertently the sample was about the 418 software, which was then up to $3.5 million, substantially more than I had claimed it would cost. A very high-priced song!

Customer Relations

The periodic meetings of UNIVAC equipment users were sometimes difficult. One such meeting drew a lot of complaints about our FORTRAN processors (for both 1107 and UNIVAC III). I was called to the meeting to hear them. Rather than attempt to answer them there, I asked them to send us all such complaints in writing, together with the source program that caused the trouble (either error or slowness). I would assign someone to run their program(s) and report to them directly on the outcome.

We received exactly two programs. One had faulty source code, which we pointed out to them. The second actually caught a small design flaw, which we fixed.

Programmer Peer Ranking

With anything over 50 technical people in your purview, it is hard to know their relative capabilities. This was particularly so at St. Paul. I conceived of a peer ranking, everyone evaluating all others. Results were a little disconcerting:

• Second-ranked was a programmer whose salary was quite low.

• Ranked last was a man with 5 children and a fairly high salary.

In the second case, I argued that the man was so ill-suited to his work that we were doing him a disservice by not prodding him to look for a position where he could really contribute and grow.

Production Control

Milton S. (Milt) Bryce was one of my direct reports. He and I got to thinking about how to make software production a proper business.

We started out simply. Each major manager was required to identify the software units he was required to produce, together with the identity of interconnecting units. An estimate was required for number of instructions and completion date. These data were put on punch cards, with space for successive (dated) estimate revisions. Also for signatures to these estimates! Managers were told that we would plot progress from these, and justification must be submitted for major deviation. They were also told that if plausible progress did not occur, they would be going back to being just programmers until they could get it right!

This was the genesis of the Software Factory, developed more fully at GE.

Professional Activities

My publication output while at UNIVAC was at the lowest rate of my career, reflecting the magnitude of the job and the amount of travel entailed (266,000 miles in less than three years).

Leaving UNIVAC

The first approach about my leaving UNIVAC came from a headhunter in New York City, on behalf of RCA, who apparently wanted me to head their programming operation. Perhaps this was because RCA, in Cherry Hill, was attempting to clone IBM systems, and I knew the IBM story. Actually they tried twice. On the first time my wife and I were given a fancy lunch at the recruiter's office. Blazing shrimp and excellent wine.

But my wife's good friend was Joan Rinaldi (secretary to Y. P. Dawkins, VP of Marketing for IBM's Eastern Region) who was dating Ed Donegan, who ran the RCA effort. From this I knew that cloning would not work for them (although Gene Amdahl managed to pull it off later). I have learned, from other memoirs of people working there then, that I chose well in refusing.

My actual departure from UNIVAC was initiated by Dr. Rader's announcement that he would be going back to General Electric, a move that he frankly admitted was due in part to his wife's wishes. Thus I was deprived of my mentor and several layers of protective armor. I then reported to Fred Raach, who was not able to withstand pressure from Lee Johnson, who had very good cause to dislike me. First Julian Goodpasture was brought in to assist me, on the theory that I was not managerial enough. Here's an example:

Dennis Hamilton once sent all software management a piece of very clever code, claiming we could not improve it. I worked on it while commuting, and after two days sent Dennis a slightly better solution, copying the same people. Apparently it was beneath managerial dignity to do so. But I always believed that managers should understand what the troops were doing, although this opinion fared no better in my future GE employment.

Around Christmas of 1964 Dr. Rader invited me to a GE management conference in Clearwater, Florida. There he introduced me to H. Brainerd Fancher, who was to run the Bull GE operation in Paris. Because Fancher was a little short on knowledge about computers per se, I was made a very good offer to help him, and you can eventually find five vignettes about that period.

Back to History Index            Back to Home Page