[ntp:questions] Re: NTP Leap Second on Windows XP

David J Taylor david-taylor at blueyonder.co.not-this-bit.nor-this-part.uk.invalid
Tue Jan 3 00:47:37 UTC 2006


Martin Burnicki wrote:
> David,
>
> David J Taylor wrote:
>> I believe that two of my Windows systems (those running Martin
>> Burnicki's code) did insert the leap second correctly, see:
>>
>>   http://www.david-taylor.myby.co.uk/ntp/ntp-events.htm
>>
>> but that confusion caused by different responses from various
>> Internet servers caused more problems.
>
> You have observed that even the servers which run our "xmas edition"
> of ntpd had offsets which exceeds the normal range.
>
> I'm assuming those offsets are due to the strong drift of some
> upstream servers. If the value in the drift file is abnormlly high
> then the system time of those servers is also drifting abnormally.
>
> If there were several upstream servers which behaved this way, with
> randomly more or less high drift, then obviously a client had
> difficulties to follow up, even if no leap second had been around.
>
> If you have a look at the event logs of the servers running the xmas
> edition, are there any entries which might be related to this? I mean
> ntpd synching first to one upstream server, then to a different one,
> etc. ("clock hopping"). Please let me know.
>
>
> Martin

Martin,

I've made some manual, edited extracts from the event log (in the absence 
of a File, Export command!), and I will send these directly to you (and 
anyone else who is interested).

Yes, there was considerable clock-hopping, and time resets with 
adjustments up to about +/- 1 second.

I have nothing but praise for your version of the software.  I perhaps 
made a mistake by not running that version on my Internet-facing servers 
(I had considered the software "beta", I should have trusted it as 
"production").  I am now using your version on /all/ my systems in 
readiness for the next leap second.

The value in my own drift file was fine, until ntpd decided to change it, 
because of the time steps ranging up to one second, hence my thread about 
drift handling needing to be more robust.

Cheers,
David 





More information about the questions mailing list