[154393] in North American Network Operators' Group
Re: F-ckin Leap Seconds, how do they work?
daemon@ATHENA.MIT.EDU (Saku Ytti)
Tue Jul 3 14:53:20 2012
Date: Tue, 3 Jul 2012 21:42:57 +0300
From: Saku Ytti <saku@ytti.fi>
To: nanog@nanog.org
In-Reply-To: <4FF33AE1.50105@foobar.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On (2012-07-03 19:33 +0100), Nick Hilliard wrote:
> Google's approach to this is interesting:
>
> > http://googleblog.blogspot.ie/2011/09/time-technology-and-leaping-seconds.html
Yes. I'm sure this is good enough for most people, most people don't need
precise time but virtually everyone needs monotonic time.
And this is easy to deploy TAI + UTC presentation using leapsecond file
lookup isn't exactly easy.
Too bad this isn't standard configuration option in NTPd.
Also one thing I wonder, why did GOOG choose to skew in just 24h, why not
the moment leap is announced? Everyone has some accuracy budget, what ever
that might be, it almost certainly is same every day. So you could live
with tighter accuracy budget the longer you spend skewing.
PC clock on average is probably like 15PPM accurate (or order magnitude of
worse, IBM servers seem to be exception). If you'd skew 3 months, your
skewing would cause inaccuracy of 0.19PPM. Skewing in single day causes
inaccuracy of 11.6PPM (still almost certainly better than free-running PC
oscillator)
--
++ytti
