Recollections of CUCC 1968-70
Some thoughts regarding the history of the CUCC [Columbia University
Computer Center] and the [IBM] 360/91....
I was a member of the first group of student consultants (September,
1968). I had been working as a gofer and informal user consultant at
Princeton's Computer Center while a high school student, and had gotten to
know the director there, Roald Buehler. He was friends with [Columbia
Computer Center founder and director] Ken King (during the 1968 garbage
strike, he sent a telegram to Dr. King, "Suggest you run for mayor
slogan garbage out garbage in"¹) and suggested to
Dr. King that students such as myself (who would be paid out of the student
aid budget) would be a good replacement for full-time CUCC staff (who were
paid out of CUCC's budget) in handling user issues. I matriculated in
September, and was installed in the user desk (in the CUCC library, next to
Nuala [Hallinan]'s desk) the following week.
My copies of the Henry Miller novels (the authoritative source on
the policies of the Cosmodaemonic Telegraph Company) are in storage, but
IIRC telegrams had a special rate if you could say it all in ten words or
less. "Suggest you run for mayor slogan garbage out garbage in" meets the
requirement. I remember spending an hour working with Dr. King on a witty
rejoinder that would meet the ten-word requirement: we couldn't come up with
The CCCP Chess Program
was part of the 360/91, and was required
for its diagnostic software. Other models of the 360 family displayed all
of their diagnostic scan information on the front panel, but the 360/91 had
enough addressable gates that the diagnostic scan was run from the 2250.
The 360/91 control panel
(which only displayed
the gates that were of interest when the machine was in 'normal' mode) was
the largest such collection of lights that IBM ever assembled -- the 'scan
to memory and display on graphic device' paradigm of the 360/91 became the
norm for future machines.
The 2250 display normally sat idle when the machine was in use. A group of
us (including Andrew Koenig, Benjamin Yalow, Aron Eisenpress, Robert Lerche,
and Steven Bellovin) discovered that GIM (IBM's painful Graphics Interface
Method) had support for the 2250, and programmed it to display a chess
board. We subsequently used that effort as the user interface for a postal
chess program originally written in PL/1 by Dr.
Hans Berliner of
Carnegie-Mellon University. This marriage
participated in the first computer chess tournament in 1970 (held at the ACM
convention downtown, under the supervision of Dr. Monty Newborne, of the
department of Electrical Engineering). I have the dubious distinction of
being one of the first people to lose a game of chess to an online chess
program (Andy, Ben, Aron, Bob and Steve all played at the master level or
better -- I didn't, and the program did). CCCP
(a pun on the Cryllic abbreviation for the official name of the Soviet
which replaced the back end with a much better set of algorithms, was a
year's worth of effort to replace the original prototype.
360/91 Loop Mode
I wrote one of the first user programs that actually went 'into'
Given that a program sequence had to be less than 63 halfwords long to be
run in loop mode, the 'loop mode' light would, at best, flicker. There was
concern at the time that 'loop mode' was a chimera, and that 'serious'
programs would not be able to take advantage of this hardware capability.
The bean counters needed proof that 'loop mode' could be activated by 'normal'
program behavior. Most observed 'loop mode' behaviour to that point was
either IBM diagnostic routines, or programs looping due to coding or
compiler error. Sans loop mode, the machine was ~1.5 times as fast as the
360/75, but cost considerably more (even before the 'education discount' and
the 'discontinued machine' discount). I was on the 'machine should stay'
side, and took an FFT algorithm from a recent Communications of the
issue, coded it in FORTRAN, and coaxed it though the FORTRAN H
compiler (an undertaking of some magnitude, as the compiler had the habits
of looping, crashing, and emitting incorrect code). Fourier transforms of
matrices of random numbers put the machine into loop mode for impressive
periods (tens of seconds) after which the program exited gracefully.
Machine Room Acreage
Somewhere on your web page
you refer to the
/91 as taking up a half-acre of raised floor. The machine room at CUCC
wasn't that big: it was all of 10K sq. ft. with support columns for Uris
Plaza running through it. Shoehorning the /75, the /91, and all of the
support peripherals into that space was a bit of an exercise. Princeton had
20K sq ft of column-free raised floor when they built their standalone PUCC
building in '68: they fitted in a /91 and had room to spare.