[ntp:questions] Losing clock ticks
stenn at ntp.isc.org
Fri Apr 13 21:11:49 UTC 2007
>>> In article <rv14f4-6tk.ln1 at gateway.py.meinberg.de>, Martin Burnicki <martin.burnicki at meinberg.de> writes:
Martin> Unlike with FreeBSD (AFAIK) there are a large number of Linux
Martin> distributions with a large number of versions each, and thus a large
Martin> number of different kernel versions are running out there, so the
Martin> probability that the ownwer of a system which encounters those
Martin> problems is higher for Linux than for FreeBSD.
I could be wrong. It just seems to me, based on the reports I see, that
there have been a few different things over the years where Linux kernels
and their distros have done things that seem to make it real difficult to
get them to keep good time. Lost interrupts in drivers, strange algorithms
in ntp kernel subroutines, different libc interfaces.
Just as another example, at one time Solaris was a great timekeeping
platform. I gather a little while ago Dave went in to the serial driver and
saw how it had been improved for certain tasks, and those improvements
really messed up the serial driver's ability to keep time well. I think
Dave sent them back a modified driver that restored the balance, but I could
be mistaken about many things in this paragraph.
More information about the questions