[ntp:questions] Re: Make NTP timestamps leap-second-neutral (like GPS time)

David L. Mills mills at udel.edu
Wed Jan 14 16:16:40 UTC 2004


I have no answer to your question for two reasons. First, I describe
only my personal observations; there may be many others with different
experiences. I watched only my personal flock, a number of servers
scattered around the world, all of which running software I verified was
identical. I can't and won't speak of other sites running j-random
versions and j-random kernels amd j-random drivers.

The second reason is that the occasion you cite is ten years old. In the
general velocity of technology change these days, occasions that old are
completely irrelevant to the events expected today. Heck, even the
memories of the last leap at the end of 1998 are in the previous
century. Tell Google to disregard anything over two weeks old.


Michel Hack wrote:
> "David L. Mills" <mills at udel.edu> wrote in message news:<4004AC3A.6248ECD6 at udel.edu>...
> > ... I can report that
> > the latest leap represented by the last second of 1998 using NTPv4 with
> > the WWV reference driver to deliver the leap warning bit and the kernel
> > leap state machine did in fact correctly insert the second as intended.
> I have no doubt about the "leap state machine" and the actual
> recognition
> of the Leap Second Event.  My question was about the stability of
> stratum-1
> servers for a certain period *after* the event, if in fact all those
> Unix
> clocks independently "absorb" the leap second.  For 1994 you described
> a
> "pinball" effect with a "tradition dating back years".  Can I assume
> that
> nothing similar has happened since then, i.e. that the timestamps
> received
> from a reputable server (stratum 1 or 2) over a good stable link show
> no discontinuity in logical value (in other words, NTP time stamps
> without
> the LSE-pending bit differ by exactly one second from the projected
> value
> of those with the LSE flag)?
> If that's the case, my primary argument for a change of protocol goes
> away.
> Michel.

More information about the questions mailing list