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

David L. Mills mills at udel.edu
Fri Oct 14 17:20:32 UTC 2005


The leap warning bit is exchanged in the NTP subnet exactly following 
announcement by NIST. The announcement can be made up to six months in 
advance of the leap event. Every NTP client and server on this planet 
and beyond should show the leap to be inserted. However, the kernel will 
not be so advised until 31 December. The kernel will implement the leap 
at minute 1440 of that day. If the kernel is not so equipped, the leap 
will not be effected until the stepout threshold has been reached, 
ordinarily after about 900 seconds of the new day. After that the time 
will be stepped one second. If the operator doesn't like the step and 
elects to slew instead, the time will not be correct until after 2000 s 
have elapsed. The reason for this behavior has been repeatedly posted to 
this group and the documents at the NTP project page.


Martin Burnicki wrote:
> Rex,
> Rex Vincere wrote:
>>Recently I have noticed that a number of servers (on and off), and
>>especially louie.udel.edu (all the time) are reporting a leap second. This
>>has been going on for a couple of weeks now.
>>This is the first time in about the 10 years I have been playing with NTP
>>that I have seen a problem like this pop up.
>>Or is this not a problem but a change to NTP I missed?
> There has been a leap second scheduled for the end of this year, i.e. to be
> introduced after 2005-12-31 23:59:59 UTC. It's the first leap second after
> 7 years. 
> However, if I understand the source code correctly, ntpd should start to
> announce the leap second insertion only one day before the time of
> insertion, so if the announcement flag has already been set this might be a
> bug.
> Martin

More information about the questions mailing list