[ale] Solved - Re: Issues with NTP - synchronizes, then falls behind...
Chris Woodfield
rekoil at semihuman.com
Sun Jul 26 12:53:17 EDT 2009
After some more poking around, managed to get this fixed, but figured
I'd share the solution...
The problem wasn't anything in ntp - it was, in fact, a crappy
internal clocksource. However, on most systems there are multiple
clock sources available, and changing them is only an echo $source > /
sysfile away.
Specifically on my box, the magic command was:
tino:~# cat /sys/devices/system/clocksource/clocksource0/
available_clocksource
tsc hpet acpi_pm jiffies
tino:~# cat /sys/devices/system/clocksource/clocksource0/
current_clocksource
tsc
tino:~# echo acpi_pm > /sys/devices/system/clocksource/clocksource0/
current_clocksource
tino:~# cat /sys/devices/system/clocksource/clocksource0/
current_clocksource
acpi_pm
Then, in /boot/grub/menu.lst, add "clocksource=acpi_pm" to your kernel
boot command args.
I picked "acpi_pm" because that seems to be the source most often
mentioned in the googled results I found when looking this up.
However, I couldn't seem to find anywhere was any sort of
documentation on what each of these clocksources actually *are*.
Anyone able to explain this further?
Thanks,
-Chris
On Jul 26, 2009, at 11:57 AM, Chris Woodfield wrote:
> Hi,
>
> Having a major problem with a server that's running Debian 4.0 - can't
> seem to keep the clock in sync.
>
> I believe that the root cause is that my motherboard's hardware clock
> is running slow, but I'm trusting that ntpd should correct for that -
> which it is not.
>
> When I start ntp up, it connects to multiple pool.ntp.org servers, and
> appears to properly synch up the internal clock. However, within a few
> minutes, that clock is several seconds behind - what's more is that
> ntp seems to detect this, as shown in the offset field of ntpq -p, but
> doesn't seem interesting in correcting it.
>
> As the diag shows below, I'm skewing a full 10 seconds within 10
> minutes - is this just too far of a delta for ntp to be able to work
> with? If so, is there a config option to resolve this?
>
> Here's a quick shell script I wrote to demonstrate the problem -
> 192.168.0.155 is a Mac that syncs to time.apple.com.
>
> --
> echo -n "Initial Localtime: "
> date
> echo -n "Initial RemoteTime: "
> ssh 192.168.0.155 date
> sudo /etc/init.d/ntp restart
> for VAR in 1 2 3 4 5
> do
> sleep 90
> ntpq -p
> echo -n "Localtime: "
> date
> echo -n "Remotetime: "
> ssh 192.168.0.155 date
> done
> --
>
> And the output - note that on first run, it goes from 6 seconds behind
> before ntp restart to ~2 seconds, but then drifts out to a full 10
> seconds behind at the end of the 450 seconds it takes the script to
> run.
>
> rekoil at tino:~$ sh ntp_run
> Initial Localtime: Sun Jul 26 11:45:34 EDT 2009
> Initial RemoteTime: Sun Jul 26 11:45:40 EDT 2009
> Stopping NTP server: ntpd.
> Starting NTP server: ntpd.
> remote refid st t when poll reach delay
> offset jitter
> =
> =
> =
> =
> =
> =
> =
> =
> ======================================================================
> host2.kingrst.c 173.14.47.149 2 u 41 64 1 50.214
> 844.062 0.001
> mainframe.cynac 216.218.254.202 2 u 27 64 1 31.627
> 1142.33 0.001
> Localtime: Sun Jul 26 11:47:13 EDT 2009
> Remotetime: Sun Jul 26 11:47:15 EDT 2009
> remote refid st t when poll reach delay
> offset jitter
> =
> =
> =
> =
> =
> =
> =
> =
> ======================================================================
> host2.kingrst.c 173.14.47.149 2 u 3 64 7 49.684
> 2195.19 1354.11
> mainframe.cynac 216.218.254.202 2 u 51 64 3 31.627
> 1142.33 1393.21
> Localtime: Sun Jul 26 11:48:43 EDT 2009
> Remotetime: Sun Jul 26 11:48:47 EDT 2009
> remote refid st t when poll reach delay
> offset jitter
> =
> =
> =
> =
> =
> =
> =
> =
> ======================================================================
> host2.kingrst.c 173.14.47.149 2 u 28 64 17 49.684
> 2195.19 1922.28
> mainframe.cynac 216.218.254.202 2 u 14 64 17 31.310
> 3886.57 1925.79
> Localtime: Sun Jul 26 11:50:13 EDT 2009
> Remotetime: Sun Jul 26 11:50:19 EDT 2009
> remote refid st t when poll reach delay
> offset jitter
> =
> =
> =
> =
> =
> =
> =
> =
> ======================================================================
> host2.kingrst.c 173.14.47.149 2 u 52 64 37 49.684
> 2195.19 2647.40
> mainframe.cynac 216.218.254.202 2 u 38 64 37 31.310
> 3886.57 2153.12
> Localtime: Sun Jul 26 11:51:43 EDT 2009
> Remotetime: Sun Jul 26 11:51:51 EDT 2009
> remote refid st t when poll reach delay
> offset jitter
> =
> =
> =
> =
> =
> =
> =
> =
> ======================================================================
> host2.kingrst.c 173.14.47.149 2 u 16 64 177 48.996
> 8992.49 5287.44
> mainframe.cynac 216.218.254.202 2 u 2 64 177 31.310
> 3886.57 3275.18
> Localtime: Sun Jul 26 11:53:14 EDT 2009
> Remotetime: Sun Jul 26 11:53:24 EDT 2009
>
> Relevant config stuff:
>
> root 22986 0.0 0.1 5492 1692 ? Ss 11:45 0:00 /usr/
> sbin/ntpd -p /var/run/ntpd.pid -u 0:0 -g
> By default Debian 5 runs ntp as UID ntp, changed to root to eliminate
> potential permissions issues...
>
> from ntp.conf:
> server 0.debian.pool.ntp.org iburst dynamic
> server 1.debian.pool.ntp.org iburst dynamic
>
> Thanks,
>
> -Chris
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://mail.ale.org/mailman/listinfo/ale
>
More information about the Ale
mailing list