On Sat, 2023-12-23 12:49:16 +0100, Jan-Benedict Glaw <jbglaw%lug-owl.de@localhost> wrote:
After swapping terminators and cables, the conclusion is that one of
the system's SCSI cable's plug doesn't make proper contact to the
cable. Using a different plug on the same cable (which isn't at the
"perfect" location though) makes it work.
So I'm now prepared with a 4000/60 :)
With the HDD image I used on the /90, that 4000/60 is running since
some hours. No network connection, no ntpd. The image is a few months
old, but for finding a misbehaving clock or lost interrupts, that
should be good enough.
While running idle (~ 3 h), I didn't notice loss of time. Maybe a
second? But no more. (And that's over 9k6 serial...)
Then I let it call gcc on a simple C file in a loop for another six
to seven hours, and now I seem to have an offset of some 1.5 to 2
seconds. No further messages (negative runtime) in `dmesg`. And a
total of 2 sec over a timespan of 9 h would be totally fine for ntpd.
I'll now run a fresh install with a newly built install ISO (already
containing the recent page invalidation patch) and give it another
try. But I wonder why that box shows a reasonable time. Maybe there's
actually an issue with the code behind adjtime()?