Re: clock suddenly slipping behind
On Fri, 27 Aug 2004, Nori Heikkinen wrote:
> on Fri, 27 Aug 2004 02:21:44PM +1000, Tim Connors insinuated:
> > Stefan O'Rear <stefanor@cox.net> said on Thu, 26 Aug 2004 20:34:17 -0700:
> > > On Thu, Aug 26, 2004 at 11:23:07PM -0400, Nori Heikkinen wrote:
> > > > over the past few days, i've noticed that my system clock gets about
> > > > ten to fifteen minutes slow over the course of a day. this is really
> > > > weird! i've been using ntpdate to synchronize it with a timeserver
> > > > whenever i notice it, and i put it in a once-a-day cron job, but i
> > > > want my system to ALWAYS be on time. i'm confused as to what's
> > > > causing this, and how i can fix it. any ideas?
> > >
> > > Perhaps your PIT is going south? (PIT = Programmable Interval Timer,
> > > a variable-frequency timer usually set to 100HZ by Linux.)
> >
> > Nope. This seriously needs investigation.
> >
> > http://www.google.com/groups?selm=2qVhI-80D-5%40gated-at.bofh.it
> >
> > The one replyer said he didn't see anythign wrong.
> >
> >
> > I had 2 machines with ntp packages and adjtimex querying two known
> > good upstreams, plus three pool.ntp.org servers, that upon upgrade of
> > sid a couple of weeks ago, broke at the rate of ~12 and ~14 seconds
> > per 10 minutes (for my two machines, very constant for each), which
> > was ~twice the rate that the OT reported). One went through a kernel
> > reboot and the other didn't, so it wasn't a new kernel issue.
> > Uninstalling ntp and adjtimex and reinstalling didn't fix.
> > Uninstalling, *purging* (so drift file and config files gone),
> > *rebooting*, and then reinstalling fixed. Doing one or the other of
> > rebooting and purging was not good enough - the kernel keeps state in
> > one case, and the ntp drift files etc keep state in the other case.
>
> wait, uninstalling what, and purging what? adjtime?
and ntp*
> > I haven't tried to reproduce this, but things to note were the drift
> > file *seemed* to have normal contents, the adjtime file was slightly
> > off (but should only affect the hardware timer anyway, and was
> > probably off because ntp was so confused - you can't calibrate the
> > hardware clock off a faulty software clock).
> >
> > One other very clued in guy on the scary devil monastery also found
> > this problem a day or two ago. I've been in communication with him,
> > and it seems these are all related. There is a hard to trigger bug
> > somewhere, but if you want to track it down, you'll prbablky need to
> > reinstall old version of ntp and/or adjtimex and just keep working
> > forwards and backwards until you trigger the bug again.
>
> right now, i've got:
>
> ii ntp 4.2.0a-11
> ii ntp-simple 4.1.0-8
> ii ntpdate 4.2.0a-11
> ii adjtimex 1.18-1.1
>
> did you get a set of versions that works for you?
Yep - I am now running
ntp: Version: 1:4.2.0a-11
ntp-server: Version: 1:4.2.0a-11
ntp-simple: Version: 1:4.2.0a-11
ntpdate: Version: 1:4.2.0a-11
and no adjtimex (now)
> and if this is a problem with these versions, should i file a bug?
So I wonder if it was because of adjtimex? BTW, the problem did originally
crop up with these versions (adjtimex was ... I don't know; it's not in
my cache anymore)
> hmm, this looks quite relevant:
> http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=265839
> (i'm also running kernel 2.4.26-1, like the submitter of this bug).
Same.
Oh well, I don't need adjtimex (what's it's purpose when you have ntp and
a hardware clock that isn't *too* bad?), so it looks like that is my
workaround.
--
TimC -- http://astronomy.swin.edu.au/staff/tconnors/
"This company performed an illegal operation but they will not be shut
down." -- Scott Harshbarger from consumer lobby group on Microsoft
Reply to: