[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

xntpd oddness



David Monro writes:
 > I'm having a strange problem with xntpd. It seems to be lying to me!
 > 
 > I'm running a ruffian under Debian and sharing filesystems with a Digital Unix
 > box. xntpd is running on both machines, and in both cases xntpdc claims that
 > they are within a few milliseconds of the time provided by the server. However,
 > 'date' indicates the machines are out by quite a few seconds, and make starts
 > complaining about modification times in the future etc.

This behaviour of xntpd is pretty annoying, it actually requires the
local clock to be quite stable, and precise within a percent fraction
of physical time. If it detects this is not the case, it still
continue to runs and compute the real time from server, but it will
not update anymore the local clock (and it does not put any warning
message in the logs).

First ensure there is no problem with the stability of your local
clock.  Several machines with pyxis chipset (miata,lx) had a very high
local-clock-unstability problem, that was corrected a few month ago by
new milos, maybe it is also the case with the ruffian.

Even if you do not have this problem, the precision required by xntp
may require you to use tickadj, if it is not sufficient, you may have
to use the adjtimex system call, (I am not sure there is already an
utility that would do instead), because on the Alpha, a timer
interrupt period is not an integer number of microseconds.

Loic


Reply to: