[ntp:questions] Re: Clock jumps 1 second every 11 minutes (Linux+SMP+NTP)

Villy Kruse vek at station02.ohout.pharmapartners.nl
Tue Aug 26 08:30:11 UTC 2003


On 25 Aug 2003 13:03:52 -0700,
    Nathan Parker <google.20.nathanpll at spamgourmet.com> wrote:


>I'm trying to syncronize the clock on a dual-processor DELL 2600 with
>3.06 GHz Intel XEONs.  It's running Linux (RH7.3) with
>2.4.18-27.7.xsmp kernel and NTP v4.1.1.
>
>Without NTP, the clock drifts 1.07 seconds slow per hour.  With NTPD
>running (to a strat 1 server) with the -x1 option, the clock JUMPS
>FORWARD one second every 11 minutes exactly -- it keeps doing this,
>even when the clock is 20+ seconds fast! Without the -x option, the
>clock still jumps forward but then NTP jerks it back again a moment
>later. I've let it run for days and it never stabilizes. This is very
>disruptive to the real-time signal processing I'm trying to do on this
>machine.
>
>I dug in the linux kernel source and found that it (apparently) sets
>the BIOS clock from the Linux RTC every 11 minutes -- is this "11
>minutes" a coincidence, or is it related?  What can I do to keep
>proper time on this machine?
>

BIOS clock same as RTC clock, the hardware clock or real time clock.
System clock, is the unix clock, which NTP works hard at keeping in
sync with the NTP server.

Supposedly the kernel should set the RTC clock from the system clock
every 11 minutes, if the kernel finds out a ntp server is active.  
The reason for that is that a NTP controlled system clock would be
more accurate than the RTC clock, and regularly resetting it will
increase the likelyhood that it is reasonably correct next time the
system is booted, which is normally the only time the value of the
RTC clock matters.


Villy



More information about the questions mailing list