[LEAPSECS] Local insertion of leap seconds

Harlan Stenn stenn at ntp.org
Fri Jan 3 16:27:31 EST 2014


Stephen,

I am not a wide-scope wizard in this area. I may not be even a
narrow-scope wizard.

Having said that, this is a problem with the timescale used by different
systems. UTC has leap seconds, but POSIX (for example) does not address
how to deal with them.

On Unix systems, NTP will apply the leap second at the rate of 1 second
per second at the end of 23:59:59. On Windows, NTP will apply the leap
second at the rate of .5 seconds/second starting at 23:59:59, which
means the leap second correction is still in-process until 00:00:01.

One can also apply this smear over longer periods, like perhaps a full
day.

The correction can be applied linearly or more smoothly:

http://googleblog.blogspot.in/2011/09/time-technology-and-leaping-seconds.html

http://stackoverflow.com/questions/11279992/math-behind-google-leap-second-smear-formula

You probably know way better than me that there is a difference between
frequency synchronization and time synchronization, and sometimes there
is a problem trying to do one with the other.

Having said this, http://nwtime.org/projects/timestamp-api/ is working
on its General Timestamp API project, to do a better (sufficient, even)
job of accurately and precisely generating timestamps. This includes
cases where the system is applying a correction, be it a general offset
correction or a leapsecond correction.

NTF needs participation and support to develop the General Timestanp API
project - I encourage all folks interested in this to help make this happen.

--
Harlan Stenn <stenn at ntp.org>
http://networktimefoundation.org - be a member!


More information about the LEAPSECS mailing list