[ntp:questions] A Suggestion For Abolishing the Leap Second
David L. Mills
mills at udel.edu
Wed Jun 6 02:36:39 UTC 2007
There are already lots of systems that do and do not respond to advance
leap notice. In the usual case of with no advance notice, the response
is to continue as before the leap and after 15 minutes make the leap. In
the older radios there was no provision for advance notice, so the radio
continued as before the leap and, typically, resynchronized correctly
several minutes later. In cases where the leap epoch is supplied
externally or by the NIST leapseconds file and when kernel provisions
are available, the kernel makes the leap as specified. However, if the
radio does not understand the leap, it will continue as above. It is
this case that the fifteen minute wait is for.
It is important that the leap be substantially complete by the end of
the inserted second, which is done by stopping or almost stopping the
clock for one second. Gradual corrections are not good unless the rate
of the correction was standardized and everybody adopted the standard.
Otherwise, a substantial difference between clock readings might occur
during the correction interval.
Rob van der Putten wrote:
> Hi thereRob,
> Wolfgang S. Rupprecht wrote:
>> I'm a firm believer in simplifying the low-level protocols. The
>> simpler they are the less likely that someone will screw them up.
>> This goes doubly for stuff that one only gets to test once every 1.5
>> years. (Personally, I wouldn't mind at all if ntp got out of the leap
>> seconds business and just distributed GPS time or something else with
>> with a consistently incrementing time scale. Make the leap seconds
>> the responsibility of some user-level code just like DST is now.)
> Whenever politicians decide to change the date on which summer time
> starts or ends, new timezone files are distributed and installed. So the
> infrastructure to deal with changes in the relationship between
> international and local time is already there.
> And if the OS already deals with translation from and to local time, why
> can't it deal with leap seconds in a similar manner. It already adds a
> number of seconds to the international time anyway.
> Having a fixed second length really is the cleanest way of doing things.
> Most applications generate timestamps, ids and filenames using some sort
> of epoch anyway, so they don't really care about 'classic' date and
> time. Just the number of seconds since some point in time will do.
> And one could run ntp on two different ports. One with and one without
> leap seconds to accommodate a gradual transition from one system to the
More information about the questions