| |
KF5JRV > TECH 02.04.16 15:44l 53 Lines 3922 Bytes #999 (0) @ USA
BID : 776_KF5JRV
Read: GUEST
Subj: Leap Seconds
Path: IW8PGT<CX2SA<N9PMO<N0KFQ<KF5JRV
Sent: 160402/1106Z 776@KF5JRV.#NWAR.AR.USA.NA BPQ1.4.65
The Internet Time Service and Leap Seconds
A leap second is announced in advanced in Bulletin C of the International Earth Rotation and
Reference Service (www.iers.org).
The leap second can be either positive or negative, although only positive leap seconds have ever
been used, and it is very unlikely that negative leap seconds will ever be required. The following
discussion describes only the insertion of a positive leap second for this reason.
The leap second is added to the last minute of the last day of a month. The event can be scheduled
for any month, but the months of June and December are preferred, and no other months have ever
been used. The leap second event is linked to the UTC time scale (not local time as with daylight
saving time), and therefore occurs at different local times in different time zones. For example, a
leap second at the end of June will occur June 30 at 5:59:59 p.m. local time in Colorado
(Mountain Daylight Saving Time, UTC-6).
The name of a positive leap second is 23:59:60, but systems that represent the current time as
the number of seconds that have elapsed since some origin (NTP, for example) generally cannot
represent that time. The next best thing is to add the extra leap second by stopping the clock for
one second at 23:59:59, and that is what the NIST time servers do. That is, they repeat the binary
time equivalent of 23:59:59 twice, and the next second is second 0 of the following day. The time
tag corresponding to23:59:59 is therefore ambiguous, since two consecutive seconds have that
name. For example, it can be difficult to establish the time-ordering of events in the vicinity of a
leap second, since the time 23:59:59.2 in the leap second occurred after 23:59:59.5 in the first
second with that name. A calculation of a time interval across the leap second has a similar
ambiguity. There are no easy solutions to these ambiguities because the format of NTP messages
does not have any means of distinguishing between the two seconds that have the same name.
There are two ways of realizing the leap second that we see as incorrect:
1) Some systems implement the leap second by repeating second 0 of the next day instead of
second 23:59:59 of the leap second day. This has the same ambiguity problem of the NIST
standard method, and also puts the extra second in the wrong day.
2) Some systems implement the leap second by a frequency adjustment that smears the leap
second out over some longer interval. This has the advantage that the clock never stops or appears
to run backward. However, it has both a time error and a frequency error with respect to legal UTC
time during the adjustment period. To make matters worse, there is no universal way of realizing
this idea, so that different systems that use this method may disagree during the adjustment period.
Both of these methods have the correct long-term behavior, of course, but neither of them is consistent
with the legal definition of UTC. Therefore, any application that requires time that is legally traceable to
national standards and uses these methods to realize the leap second, will have a time error on the
order of 0.5 - 1 s in the vicinity of the leap second event.
All NIST time services provide some advance notice of the leap second, but the details vary from one
service to another. For example, the NIST digital telephone service (ACTS) provides advance notice
from the start of the month in which the leap second will occur. The NIST NTP servers provide advance
notice starting from 00:00 UTC on the last day of the month when the leap second will occur.
Most versions of UNIX (and its derivatives, such as Linux, FreeBSD, ...) have support for the leap
second built into the operating system. Many desktop systems do not have any native support at all
for leap seconds, although there are some third-party applications that do this.
Read previous mail | Read next mail
| |
