[ntp:questions] Re: ntp servers reporting leap second erroneously?

David L. Mills mills at udel.edu
Wed Oct 26 00:11:16 UTC 2005


Martin,

As a practical matter, the poll intervals and for that matter the 
seconds interrupt used by the reference clock drivers are not 
synchronized between systems, only the system time. In principal it 
would be possible for each client to calculate how many ticks to the 
appointed event, but not simple at all in the context of the current NTP 
engineering model.

It may in fact be easier to have a separate process whose only mission 
to step the clock back one second at a designated time and activated by 
a cron job or whatever the equivalent in Windows. Don't do a slew at 
this point; that would drive the NTP clock discipline nuts.

Dave

Martin Burnicki wrote:

> Terje,
> 
> Terje Mathisen wrote:
> 
>>Martin Burnicki wrote:
>>
>>
>>>Under Linux, the leap second announcement is passed to the kernel which
>>>handles it properly.
>>>
>>>Under Windows, the time is stepped back by one second a few minutes after
>>>the time of leap second insertion.
>>
>>If nobody else volunteers I'll take a look at the Win source code and
>>see if I can figure out a reasonably good way to insert/delete a second
>>by an extra clock slew that avoids resetting the clock drift value.
> 
> 
> The problem that the time is stepped back by one second a few minutes after
> the time of leap second insertion should be the same for all operating
> systems which have a sytem clock which is not aware of leap seconds.
> 
> So I think the preferred way to solve this would be to add this feature to
> the global code which is compiled on such systems. That piece of code is
> managed by Dave, so I'd primarily appreciate if Dave would implement this.
> 
> Dave, is there a chance you could do this? I'm sure it would be appreciated
> by lots of people.
> 
> 
>>>As I've reported earlier on this news group even the Windows w32time
>>>service is capable to slew the Windows system clock at UTC midnight to
>>>compensate the leap second offset in just a few seconds.
>>
>>Which is why I'm willing to dig in. :-)
> 
> 
> If Dave is not willing to implement this it would be great if you could dig
> in to implement it as a Windows-specific workaround. 
> 
> 
>>"Anything Microsoft can do, we can do better."
> 
> 
> That should be our goal ;-)
> 
> There are lots of instances of ntpd running under Windows today. In this
> newsgroup we keep telling users that ntpd for Windows works better than
> w32time. Should we accept now that w32time, which has not been aware of
> leap seconds at all for a long time, can handle leap seconds much more
> gracefully than ntpd?
> 
> Ntpd simply steps time back and we all know that this is not a good thing.
> 
> Anyway, if you or Dave would implement a way to handle leap seconds better
> on systems like Windows, and if you don't have the chance to test the code,
> I'll be happy to do that using the simulator I have here.
> 
> 
> Martin




More information about the questions mailing list