[ntp:questions] Some issues about NTP ( Server 2008R2)
David Woolley
david at ex.djwhome.demon.invalid
Mon Jun 2 17:24:47 UTC 2014
On 02/06/14 15:50, hschulla at gmail.com wrote:
>
> Some days ago I tried Meinberg NTP on my Server2008R2.
There isn't a Meinberg NTP as software for Windows, there is only the
University of Kansas software with a Windows installer from Meinberg.
> But after a while I recognized, that Windows-Time was not set when polled.
> Polling was at 1024 (17min) and time was set only in 90-120min intervals.
> Here a short copy of the system log:
>
> Die Systemzeit wurde von 2014-06-01T17:49:58.337959600Z auf 2014-06-01T17:49:58.154000000Z geändert.
If ntpd sets the time at all, except once at startup, it is in
difficulties. ntpd sets the frequency, when operating normally.
I think you may have overridden the minimum polling interval, as ntpd
should have ramped down to the minimum if it is stepping.
When ntpd steps, it needs to have enough samples showing a problem
first, then it waits for about 15 minutes to make sure that the problem
is real and finally, I think it needs to refill the 8 entry minimum
delay filter.
When ntpd is not stepping, it oversamples (in terms of the desired
filter bandwidth) by more than a factor of 8, then only uses the sample
from the last eight with the lowest delay, as that is assumed to have
the least error due to path asymmetry.
All the steps seem to be backwards, which probably means that your
motherboard clock is is more than 500ppm fast. Although this is best
treated as broken hardware, and the motherboard replaced, on Unix and
Linux you can correct for it in units of 100ppm. I'm not sure that
there is any equivalent tactic available on Windows.
Repeated steps can also happen if something else is disciplining the
clock at the same time as ntpd.
Overriding minpoll and maxpoll can sometimes make it difficult, or
impossible, for ntpd to lock onto the correct frequency, so it is just
possible that that would be sufficient to cause the stepping.
Basically you are seeing too many of these messages not too few.
More information about the questions
mailing list