Anyway, it periodically restarts itself, and a couple weeks ago it
shut itself down and didn't restart. I think I remember reading about
this, but I don't recall where -- either it's the G4 (Heat? There's
and extra pair of drives in there) or X Server or Retrospect (bug in
that version?).
Let me guess, you have a SCSI based tape backup?
Apple changed the SCSI system around 10.2.6 or so (Don't remember off
the top of my head.) and it can cause non-HD SCSI devices to hang or
stall. Retrospect is often blamed for this. What happens is the
kernal_task hits the roof and because it's the kernel it won't yield
time so that watchdog can reset the PMU doomsday timer. You can turn
off the "Restart on freeze..." setting in Energy Saver.
What I do to look for this is to open a top -s2 -u window and then fire
up Retrospect. Open a second window and have a kill -9 <retrospect's
PID> waiting and ready; if you are quick enough you can catch it an
kill the SCSI using process to free the system. See if putting the tape
in motion by opening the Configure > Devices window causes the
kernal_task to shoot through the roof, meaning 100% or more of your
available CPU. I've seen a single CPU G4 report kernal_task usage of
140% or so before.
Make sure retro is all the way up to date. Make sure your ATTO drives
and firmware are all the way up to date. The OS bundled ATTO drives are
way out of date, although sometimes just re-installing the OEM drivers
will get you through a single backup. (Which, makes no sense, but
sometimes happens.) You might need to set the speed and sync rate on
the SCSI card by hand too. ATTO has info on all of this on their site.
Dantz does not.
