FreeBSD not stable enough for Xen environments?

Jeremy Chadwick koitsu at FreeBSD.org
Mon Nov 17 12:08:50 PST 2008


On Mon, Nov 17, 2008 at 03:01:02PM -0500, N.J. Thomas wrote:
> * Maxim Khitrov <mkhitrov at gmail.com> [2008-11-17 14:47:00+0000]:
> > > I've not seen any problems with the clock on my RootBSD Xen system.
> > > I do run the ntpd in base and on average, my clock is usually only
> > > about 15ms away from "true UTC".
> > 
> > That's interesting. Can you post your `ntpq -p` output here?
> 
> Sure:
> 
>     $ ntpq -p
>          remote           refid      st t when poll reach   delay   offset  jitter
>     ==============================================================================
>     +clock.trit.net  192.12.19.20     2 u  529 1024  377   81.554    2.870   6.477
>     +mail.honeycomb. 192.43.244.18    2 u  408 1024  377   44.091   10.986   8.250
>     *tuppy.intrepidh 64.142.103.194   2 u  413 1024  377   67.709   15.626  10.327
>     +clock3.redhat.c 66.187.233.4     2 u  445 1024  377  147.283   24.455   9.397
>     +204.34.198.40   .USNO.           1 u  409 1024  377   88.746   20.620  10.405
>     +tick.usno.navy. .USNO.           1 u  427 1024  377   20.848   18.916   8.212
>     +ntp-s1.cise.ufl .GPS.            1 u  421 1024  377   45.709   18.067   9.222
>      LOCAL(0)        LOCAL(0)        10 l   18   64  377    0.000    0.000   0.004
> 
> This is what I pretty much used to eyeball my offset earlier.
> 
> > When ntpd is running, its polling interval stays very low (around 64
> > seconds) because it keeps having to reset the clock. My message log is
> > filled with the following:
> 
> Intersting, I see the same in my logs, but the frequency seems to be
> much less than yours, e.g. for the month of November:
> 
>     Nov  1 00:08:22 zaph ntpd[678]: time reset -0.129649 s
>     Nov  3 15:33:09 zaph ntpd[678]: time reset -0.137509 s
>     Nov  4 03:11:51 zaph ntpd[678]: time reset +0.237734 s
>     Nov  4 03:34:23 zaph ntpd[678]: time reset -0.150326 s
>     Nov  4 13:05:20 zaph ntpd[678]: time reset +0.317738 s
>     Nov  4 13:32:06 zaph ntpd[678]: time reset -0.560629 s
>     Nov  4 13:54:35 zaph ntpd[678]: time reset +0.265391 s
>     Nov  4 15:43:55 zaph ntpd[678]: time reset -0.163660 s
>     Nov  7 17:31:03 zaph ntpd[678]: time reset -0.130039 s
>     Nov 10 18:29:19 zaph ntpd[678]: time reset +0.169785 s
>     Nov 10 19:46:26 zaph ntpd[678]: time reset -0.146554 s
>     Nov 10 20:27:08 zaph ntpd[678]: time reset +0.891811 s
>     Nov 10 20:53:59 zaph ntpd[678]: time reset -0.774636 s
>     Nov 10 21:35:45 zaph ntpd[678]: time reset +0.384227 s
>     Nov 10 22:33:46 zaph ntpd[678]: time reset -0.194131 s
>     Nov 11 12:34:25 zaph ntpd[678]: time reset +0.433002 s
>     Nov 11 13:01:09 zaph ntpd[678]: time reset -0.335592 s
>     Nov 11 15:17:45 zaph ntpd[678]: time reset +0.933537 s
>     Nov 11 16:01:42 zaph ntpd[678]: time reset -0.510371 s
>     Nov 11 17:29:41 zaph ntpd[678]: time reset -0.133244 s
>     Nov 11 19:16:41 zaph ntpd[678]: time reset +0.191431 s
>     Nov 11 19:42:30 zaph ntpd[678]: time reset -0.458738 s
>     Nov 11 20:09:16 zaph ntpd[678]: time reset +0.207999 s
>     Nov 11 20:36:06 zaph ntpd[678]: time reset +0.143897 s
>     Nov 14 01:29:44 zaph ntpd[678]: time reset +0.134492 s
>     Nov 15 13:13:36 zaph ntpd[678]: time reset +0.199937 s
>     Nov 15 14:45:09 zaph ntpd[678]: time reset -0.205131 s

What time counter source does this box have available?  The following
will list what's being used (hardware) and what's available (choice):

sysctl kern.timecounter.choice
sysctl kern.timecounter.hardware

Other ideas:

Look into the "fudge" operator of ntp.conf.

Try deleting your ntp driftfile.  Note that if you do this, it will take
a day or two for things to "level out".  It tries to figure out the
"average" skew rate your system clock has.

> > And so on... Could it be a problem with the hardware on host machine?
> > I use the same ntp.conf file on several FreeBSD 7.1 servers, and the
> > VPS is the only one that has this problem.
> 
> I checked on my other FreeBSD boxes (all 7.0) and none of them (VPS or
> otherwise) exihibit this problem.

Then there's a very good possibility it's hardware-related.  At my
workplace, we've had two separate machines in the past couple months
had clocks which "went crazy" -- ntpd reporting 4-5 seconds of skew
every 25-30 minutes.  In both cases, the problem turned out to be
broken/bad hardware (crystal or TSC gone bad).

Just something to keep in mind.  :-)

-- 
| Jeremy Chadwick                                jdc at parodius.com |
| Parodius Networking                       http://www.parodius.com/ |
| UNIX Systems Administrator                  Mountain View, CA, USA |
| Making life hard for others since 1977.              PGP: 4BD6C0CB |



More information about the freebsd-questions mailing list