journal of irreproducable results (indirect Audi content)

[Mike Arman]

Ghosts in the manchine . . . thinks that go bzzt-phttt-poof in the car, and
all the smoke escapes, and you can never find out where.

You are in a maze of twisting passages, all different, yet strangely all
the same (That one's easy - it's the wiring diagram of any type 44!)

And now the story - true, by the way.

Way back in the dawn of computers, a friend of mine was project engineer at
GE for OMIBAC - early 1950's (prehistoric) computer which used several
thousand vacuum tubes - 6SN7s if anyone remembers them.

Problem was that every time they turned it on, several tubes blew, and then
they spent their mornings looking for the dead tubes (6SN7 had a metal
case, so you can't just look at it to see if the filament is lit - you have
to TOUCH each one to find the cold ones . . .) so they could do some work
by the afternoon.

They found that reliability was increased simply by leaving the filaments
on all the time, and shutting off the B+ voltage. "Hey guys, we're onto
something here!" Reliability was increased further by leaving the high
voltage on all the time, too. Then the leap of logic: Lets give it
something to do at night while we're not here - the electronic equivalent
of twiddling your thumbs! Quick program, divide by zero, and the machine
happily runs all night, not knowing it can't ever reach the end of the
calculation.

Only problem - next morning, the machine has an answer! It is wrong, but
that says something is wrong with the hardware because it HAS divided by
zero and gotten an answer . . .

Massive diagnostic effort, machine checks out just fine, come back next
day, same result - garbage!

Weeks later, they found the problem. Turns out there is a rack of neon
lights in part of the computer (D to A converter, or something), and the
ionization potential of a neon light ("firing voltage") varies with the
ambient light. Each night, the engineers turned off the room lights and
went home, and the ionization potential changed - instant garbage! In the
morning, the room lights went back on, the ionization potential changed
BACK, and of course everything worked OK again.


Final Diagnosis: "The computer is afraid of the dark."

Cure: Leave the room lights on all the time. Machine worked fine, happily
ever after.

I heard this from the guy who led the project . . . a few years later he
helped design the on-board command computer for Apollo.

Best Regards,

Mike Arman