IP: 34 years ago!

[Dave Farber <farber () cis upenn edu>]

From: Robert Rosin <rosin () shell monmouth com
Message-Id: <199911031435.JAA13723 () shell monmouth com
To: farber () cis upenn edu

The latest issue of the IEEE Annals of the History of Computing included
a "biography" by a man named Glen Meyers about his experiences as
an IBM field engineer in the 50s-80s. It brought back several
related memories, including the visit of the Snobol team to Yale
in 1965 - almost 35 years ago.

I shared my recollection of that visit with Meyers and with the Annals
editor-in-chief, who sent it to his "Anecdotes" editor for consideration.
I thought you might enjoy reading what I wrote.

By the way, I still see Ivan now and then in Red Bank, most recently
at a chamber music concert. He ran for another term on the Red Bank
council today. Don't know whether he won.

Bob
---------------------------------------------------------------------------
  ////// ////// ////// Robert F. Rosin
  // // // // // 23 Haddon Park
  ////// //// ////// Fair Haven, NJ 07704
  // // // // // Voice/Fax: 732-747-7019
// // // // // E-mail: bob () rosin net Web: rosin.net
---------------------------------------------------------------------------

  Software Diagnosis of Backpanel Wiring Problem

In 1964-5 I was assistant professor of engineering and applied science
at Yale and also served on the Computer Center Staff. In this role,
I decided to install an interpreter for the Snobol programming language
on the Yale academic computer system, which was an IBM 7094/7040 Direct
Couple System. In this configuration, the 7040 served as a "spooling"
card reader and printer for the 7094, to which it was interfaced as a
virtual device through I/O channel E. In this configuration, the system
provided excellent turn-around time, especially when compared with the
more prevalent batch systems of the day.

The 94/40 had been installed in summer of 1964, shortly before I came to Yale.
Shortly after arriving, I installed MAMOS - a version of the University of
Michigan Operating System developed at the University of Maryland -
on the 94/40 as a subsystem under IBSYS/IBJOB.

Snobol, developed at Bell Laboratories by Ralph Griswold, Dave Farber,
and Ivan Polonsky, was a string manipulation language whose intrepreter
was remarkably system independent, especially in those days. In installed
Snobol in the late fall and ran a couple of simple test programs. But
the second or third program, three or four lines of code, produced completely
incorrect results.

A month or two later I found the time to reinstall Snobol directly under
IBJOB, but with the same results. I tried a third time, and the interpreter
still failed. I had been in touch with the Snobol authors during this period,
even sending them a few voluminous dumps at their request. They saw nothing
obvious, although I did receive a mysterious comment about a "fence"
in one note. Finally, Ralph, Dave, and Ivan decided to drive up to
New Haven for a day and try to solve the problem on site.

They showed up around 9:00am, and Ralph asked me to run a simple program.
With 10 minute turn around time (and staff privileges), we quickly saw
that the program worked correctly. However, when we then tried my simple
test program, we observed the usual consistently incorrect results.

The Snobol authors spent much of the day running short programs
and examining forced core dumps. Finding no software errors, Dave
Farber began to believe that the 7094 had a hardware problem.
This seemed unlikely to me; after all, the 7094 had been running
hundreds of jobs a day for over 6 months with no apparent errors.
Occasionally a large, complex Fortran program might have to be rerun,
but that would be expected in any computer system of the 60s.

Late in the afternoon, Farber (who, by the way, is now an international
computer networking guru at the University of Pennsylvania) stated that
there was a problem with index register 7 that maintained the "fence"
used to limit the length of a string matching scan. He said that if
he could run the 7094 uncoupled from the 7040 for a few minutes, he
could locate the problem.

We got permission from the director of the Computer Center to "dry up"
the job queue and run a test. We powered down the 7040, and Farber
keyed in a 8-10 instruction program loop from the console of the 7094
and pressed "start". In an instant the 94 halted with a channel trap
from channel E, to which the 40 was normally connected. He repeated
this anomaly a couple of times, then said, "Just give this program
to your CE's - they will find a hardware problem. If you beat on
index register 7 long enough and fast enough, it generates a trap
on channel E."

Sure enough, the following week it was discovered that there was cross
talk in the back panel wiring between index register 7 and the circuitry
added to channel E to support Direct Couple operation. In retrospect,
this also helped explain those occasional Fortran failures.


_____________________________________________________________________
David Farber
The Alfred Fitler Moore Professor of Telecommunication Systems
University of Pennsylvania
Home Page: http://www.cis.upenn.edu/~farber
Email: farber () cis upenn edu

Home: +1 610 274 8292; Cell and Office: +1 215 327 8756; Fax:  +1 408 490 2720