return to main 1401 Restoration Page
Fran Underwood - 1401 - Design Philosophy
Fran Underwood was called the 1401's "Chief Architect".
Fran likes to point out that he was also the only architect. ;-))
- Fran Underwood to Robert Garner - March 2, 2011
- Robert Garner and Fran Underwood - March 4, 2011
- Charles Branscomd [1401 Project Manager] to Robert Garner - March 6, 2011
Fran Underwood to Robert Garner - March 2, 2011
I want to clarify a couple of points, although they have been discussed ad nausium. Consider how Unit Record went about setting up a plugboard: After gaining a good concept of the problem, they had to describe the process by writing down, in some form, the card columns to be read, which were to be printed by which type bars, which counters were to be employed, and how the results were to be presented, in print or punched cards.
I don't recall that there was any formal method provided by IBM to help in this task. The final document was a wiring diagram of the plugboard used to accomplish the task. IBM supplied a pad of plugboard layouts that the customer used to draw the wiring. Note that this was all accomplished by hand (and trial-and-error) without any automation aids. This process and its associated expense had to be repeated for every task.
I envisioned my job to be one which eliminated the awkward, expensive plugboard, bring some order into the job description process, retain the basic concepts of UR processing (contiguous columns, alpha-numeric language, etc.) and some automation and some sorely-needed versatility. I accomplished all this with the help of the superb talent of highly-dedicated people.
Word Marks have received an interesting amount of comment along the way. I saw word marks as a nearly ideal way to cope with the problem of contiguous fields, and then set about building a case for them as far as cost was concerned. In retrospect, I did the right thing.
My first few months on the job ( as 'Chief Architect' :-) ) I had the simplistic view that the Machine Language was 'Alphanumeric codes with decimal addressing and arithmetic', and that was all that was necessary. But then I came to realize that there was very great potential in high-level languages with their assemblers and compilers. I take no credit for their part in the success of the 1401.
Another comment: Re: speed. Unit record machines were big, heavy and slow (comparatively). They all worked on a 'machine cycle' geared to the speed with which cards could be read and lines could be printed. The 'active logic elements' were electomechanical relays with operating times in the order of 10-20 milliseconds. At these speeds, it was natural that data flow in parallel if work was to be accomplished in reasonable time. This required many duplicated circuits. With the advent of electronics (particularly Transistors), the operation time of the logic elements increased one-thousand fold (10-20 microseconds). With this blazing speed available, we could now eliminate those costly duplicate circuits by processing data in a serial-by-character fashion, thus using one circuit repeatedly. Remember, we were still tied to the limiting cyclic nature of input/output devices. Nanosecond logic would not have been a copasetic implementation at that time.
For whatever its worth....
Robert Garner and Fran Underwood March 4, 2011
e-mails between Fran Underwood and Robert Garner
THANK YOU for your write-up describing your thoughts behind your design of the 1401! This is invaluable (esp for us folks who never lived in UR land and barely comprehend it now.) I'm cc'ing/sharing your thoughts with a few other folks from the era to invite any additional thoughts or comments they might have.
I envisioned my job to be one which eliminated the awkward, expensive plugboard, bring some order into the job description process, retain the basic concepts of UR processing (contiguous columns, alpha-numeric language, etc.) and some automation and some sorely-needed versatility.
Do you recall if your "control panels are bad" mind set had been formulated while you were in ASDD?
As I have said before, you had to be there to appreciate the level of technology and the mind-set in order to understand the environment. There was no general acknowledgement that 'control panels are bad'. Lots of people didn't like them; they even hated them. Some didn't even like punch-cards! But, that was the way that it had always been done, and no alternative was offered! What opened my mind was the inordinate cost of control panel implementation in WWAM. I immediately saw a way out: eliminate control panels! The rest is history.
I assume your predisposition toward stored program computers was a much larger motivator than say the discovery (by you and/or Ralph Mork) that the WWAM's transistor interface to its control panel was too costly (about a third of the cost of its solid-state electronics, even though the WWAM already had nicely designed efficient byte-serial and memory-to-memory data paths)? It strikes me that the observation that the WWAM's interface to its control panel was too expensive was an mainly an expedient fact that no one could really argue with (i.e., it put a nail in the coffin of plug boards for the "transistor age," avoiding a religious fight over their efficacy?)
I saw word marks as a nearly ideal way to cope with the problem of contiguous fields, and then set about building a case for them as far as cost was concerned. In retrospect, I did the right thing.
I agree. Even though you may not have realized it at the time, the arguments that word marks egregiously consumed 1/8 of memory quickly became irrelevant, given that software itself, try as programmers might (or not ;-), became much more "inefficient" (as programs grew in size, compiler inefficiency came into play, etc.). Also, I've noticed that entire planes in the core stack in the main frame are used for the 1401 card buffers and the 1403 print buffers, "wasting" many more core locations in a 2K or 4K configuration.
Do you recall how you defended your use of word marks at the time??
The 'active logic elements' were electomechanical relays with operating times in the order of 10-20 milliseconds. At these speeds, it was natural that data flow in parallel if work was to be accomplished in reasonable time. This required many duplicated circuits.
Thanks for this natural insight. Although it stares you in the face (i.e., many parallel plug wires, and an article I should forward to you where a last hold out argued that control panels were better than stored program machines - there are skeptics for every paradigm change ;-), I hadn't thought of UR machines as parallel processing entities. I assume the "duplicated circuits" were mainly counters (and registers)? A silly question: What other circuits were naturally duplicated? Certainly for multiplication and division one needed a 603/604.
Other duplicated circuits were " class selectors', 'field selectors', etc. These consisted of large bank of relays that had multiple transfer points. all of which were brought out to the plugboard. There were no such things called 'registers'. I'm talking about Accounting machines, here: 402, 403, 405, 407.
For whatever its worth....
A lot, thanks!
Charles Branscomd [1401 Project Manager] to Robert Garner - March 6, 2011
One thing that hasnít been mentioned about plugboards Ė for each application (payroll, billing, inventory control, etc.) you need at least three plugboards Ė one each for the reproducer, the collator, and the accounting machine. The reproducer plugboard is rather simple but those for the collator and accounting machine can range from fairly simple to very complex. I donít know whether Franís experience is similar to mine but many of the most complex plugboards I wired was for the collator. The interpreter also requires a simple plugboard when you want to print information in the card onto the face of the card.
I was a fan of word marks from the time of my first exposure to them because of their simplicity and efficiency. But we did have some people who felt strongly that field length designation should not be contained in the data fields. Iím sure the San Jose people who proposed a 305 follow-on instead of the 1401 made this point since the 305 designated field length as part of the instruction. Fortunately this never became a big issue, partly because a member of HQS. staff Ė a Mr. Macpherson Ė was a very strong supporter of word marks. I donít know how much interaction Fran had with Macpherson (I donít remember his first name).
Fred Brooks [famed author of Mythical Man Month] expands:
John MacPherson was an IBM vp in various engineering and tech roles. A wonderfully warm and wise man, not a very aggressive manager.
I quite agree with Chuck on the plugboards, especially that of the 077 collator, and the 402 & 407 accounting machines. The 602A multiplier was a beast, too, if I remember right.
Image of 602 w plugboard Columbia University