🚀 go-pugleaf

Someone posted a note about running PDP-10 diagnostics with the room
lights out.  Running the shift and rotate diagnostics with the "Shift
Counter Maintenance" switch set to pause between each shift was good,
but only exercised the CPU and a little memory.

Before DEC wrote some system-wide diagnostics, we were having occasional
problems with our Ampex core memory.  Memory diagnostics didn't find the
problem, so I was asked to write something better.  I came up with something
that ran as three standalone processes.  The highest priority process
read and wrote stuff to the RM10B (Bryant drum), the fastest rotating
memory we had.  The next did the same to a RP03 (30 MB washing machine).
The last just checked memory ala a standard memory diagnostic.  Each process
stepped through a set of patterns and wandered through memory.  Each I/O
took a reasonable amount of time.

So, when it was running:
> The RC10 (RM10B controller) and its DF10 (data channel) showed the machine
  running through each pattern.  The DF10 had displays for both memory address
  and data, I forget about the RC10, but it looked busy.
> The RP10 (RP03 controller) and its DF10 did similar stuff, except slower
  and with occasional pauses when the RC10 needed servicing.
> The memory boxes (lights for address and data) showed the effects of zero
  to three simultaneous users.  Interleaving smeared some of the activity,
  but the simultaneously busy effect was good.
> The poor processor was very busy jumping back and forth between the three
  processes.  One pattern would appear for an instant, to be replaced by
  another, then the original one would reappear.  You could almost see the
  machine sweat.  Even the Interrupt-in-Progress lights looked busy, and
  they were positively boring when the OS was running.

The effect with room lights off was even more impressive.  I had always felt
a good diagnostic looked interesting, and was quite pleased with my creation,
even though I had no intention of making such a show when I started.  And yes,
it even found memory errors, although by that time we had concluded that
turbulance was wearing out the insulation on the wiring in the core planes.
(The occasional plume of smoke from a pulse transformer was a related clue.)
The Ampex memories were actually 40 bits wide, the PDP-10 memory was just one
of their uses for it.  We just started replacing core planes with the spare
bits, after too many failures we shiped it back to Ampex for repair.

Then there was the prototype RCA memory, which was Really Crappy Aparatus.
But pretty, with red, green, and, white lights.  Very Christmasy.  We
generally wiggled ICs in the crappy sockets RCA used until the beast
worked, and then ran with it until a couple crashes later (like around 2 PM).
Then we'd give up and disable it for the rest of the day.  Too bad - it was
64K words (288K bytes) and fast (only 1.0 usec instead of the 1.65 on the
MB10s.)
-- 
| A pride of lions               | Eric J Werme                   |
| An odd lot of programmers      | uucp: mit-eddie!alliant!werme  |
| A Constitution of Libertarians | Phone: 508-486-1214            |