22

would run his program, and while he was figuring out how to correct the inevitable errors another would be using the machine.

The lone computer operator turned the machine over to us and retired to a warm office. The process was to read some cards that started the FORTRAN compiler, then to read a deck of cards containing the program. A horrid old 100 lines-per-minute printer would produce a listing; and if the program wasn’t too bad the machine would eventually punch out a deck of cards containing the executable program. Then you put those cards in the reader and started the machine again. All this was accompanied by much spinning of reels of tape.

One of the engineers was quite knowledgeable about the computer. He would sit at the console and now and then stop the machine and examine something in the big panel of lights; or maybe he would manipulate the switches to alter the value of a variable or part of the program. The rest had to make do with what the printouts told them.

After a few hours of this, during which I closely examined a lot of the hardware to see how it was built, one of the engineers concluded that the machine was malfunctioning. He called in the operator, who ran a test program that confirmed something was wrong. The operator said he would notify the IBM field engineer when he came in in the morning, and we all went home to bed. “Some nights you get your work done, some you don’t,” was what they told me.

A little later the 704 was replaced by a more modern IBM 7090 transistorized computer. I mentioned earlier that in those days transistors were fast enough for use in Teletype equipment but were too slow for computers. In the 7090 IBM had got around the problem by using some special kind of “drift” transistors and emitter-coupled logic. The transistors were fast (well, 60 MHz was fast in those days) only if the collector-to-base voltage was about 6 volts. To get around the voltage offset from input to output there were circuits with NPN transistors and one set of logic voltage levels, and others with PNP transistors and a different set of levels.

Logic designers had to alternate stages of circuits of the two kinds. It took quite some time for IBM to get the 7090 all installed and working, perhaps a couple of months, so I got a good look at the innards of the machine before it was closed up and the machine room was closed to visitors. It was built out of the IBM SMS circuit cards, slightly larger than those used by Teletype and having thinner lines and more edge connector fingers.

These cards were assembled by automatic machinery. The backpanels were all wire-wrapped automatically by Gardner-Denver machines. I later learned of something really weird about the 7090. It used an eight-phase clock. A cable carrying the clock signals between two cabinets had as much time delay as one phase of the clock. The wire labelled “phase 1” at the sending end of the cable was labelled “phase 2” at the other end

23

because by the time the phase 1 pulse got through the cable the time was phase 2 at the sending end. This could have driven someone batty, a cable with different names at the ends of a piece of wire.

For some reason the 7090 systems employed the same lousy card machines and line printers that were supplied with the 704. These were never used in normal operation. Instead every 7090 site had at least one 1401, a small computer which was used as a card-to-tape and tape-to-printer converter.

Card decks were fed into the 1401 and written to tape; then the tape was manually moved to the 7090. A small monitor program in the 7090 read the tape and ran the jobs one after another. Output went to another tape drive. When the output tape filled up it would be carried to the 1401 for printing. The 1401 had really nice peripherals, especially the 1403 printer. This ran at 600 lines per minute and used a print chain rather than a drum. It was very impressive to watch; and it made an impressive amount of noise.

It was really a new generation of printing technology. In drum printers the same character is present at all print hammer positions at a given time. In chain printers there is a different character at each hammer position. Hence it was necessary to scan all the characters in the chain and compare with all the characters to be printed before each hammer strike time.

24

count because the collector pullup resistor of one stage is also the base pulldown resistor for the following stage. Each transistor stage was a voltage inverter but was not considered to invert the signal logically. A logic inverter was an emitter follower, so it didn’t invert voltage.

For their transistor machines IBM had developed an elaborate design automation software system. Engineers drew logic diagrams on special sheets, which were then keypunched by clerks. The software worked out the wiring of the machine. It had a data base of all the different circuit cards so it could check for things like overloaded signals and incompatible connections.

When all the errors were got out it would generate a card deck of instructions for a Gardner-Denver wiring machine and print logic diagrams on a line printer using a special font that included vertical and horizontal lines and corners of boxes. Aside from the saving in drafting labor this practically guaranteed that the logic diagram agreed with the way the machine was actually wired. Logic signals could have fairly long names, enough to make the logic diagrams largely self-explanatory and reduce the amount of written explanation of how the logic worked. It was quite an impressive accomplishment, particularly in an era when a large computer had 4096 36-bit words of memory.

My engineer friends complained about the IBM 7090 operation; they were no longer allowed to come in late at night and have hands-on access to the machine. They had to leave a card deck in the evening and pick up the results the next morning. The slightest mistake in the cards meant waiting another day for results.

One group of engineers on the base had their own IBM 1620 computer. This was a small slow machine. On this particular one the only printer was an IBM electric typewriter, so they had to learn to write programs that solved their problems without producing much output. Edwards AFB at that time also had a large punched card accounting operation using electromechanical card machinery.