[time-nuts] FreeBSD 7 ntp server

[M. Warner Losh]

In message: <1231b6a80812312047k5d2d702djcdd9610f47bc945a at mail.gmail.com>
            "Steve Rooke" <sar10538 at gmail.com> writes:
: 2009/1/1 M. Warner Losh <imp at bsdimp.com>:
: > In message: <f6bol4dpdl7b5oh32h3mo18pfki549hboh at 4ax.com>
: >            Neon John <jgd at johngsbbq.com> writes:
: > : On Wed, 31 Dec 2008 17:57:33 -0700 (MST), "M. Warner Losh" <imp at bsdimp.com>
: > : wrote:
: > :
: > : >In message: <b3bd5fcb0812311604q2c8f7129v9b1185457c8a93d9 at mail.gmail.com>
: > : >            "Robert Darlington" <rdarlington at gmail.com> writes:
: > : >: Okay, not very fun.  I was hoping to see ...58,59,60,00.  Instead my
: > : >: system ticked :59 twice.    Here's the output of my not so very
: > : >: scientific logs (up arrow, enter, over and over):
: > : >
: > : >That's the correct output.  It isn't possible to tick 60 with a POSIX
: > : >time_t, so second 59 is replayed so that we don't cross a day
: > : >boundary.
: > : >
: > : >Warner
: > : >
: > :
: > : I wonder how application software handled that.  Say, a transaction processing
: > : machine handling a few thousand transactions a second where the time stamp
: > : matters.  What did the high res timer do?
: >
: > Same thing it normally does...
: >
: > : I'm thinking about, for example, stock trading where the first bid wins.
: > : Sub-second resolution is needed there, I think.
: >
: > That's one of many reasons why I think that leap seconds are
: > fundamentally incompatible with POSIX.
: >
: > : I wonder if this was a mini-Y2K and folks haven't realized it yet?
: 
: Seems to have worked perfectly under OpenSUSE 11.1, kernel 2.6.27,
: with NTP here. It's just the poor Windblows systems that I worry
: about.

Actually, most of the effects on systems that use an absolute time for
timeouts.  posix threads can cause a stutter of 1s.  This can be quite
hard to detect in many systems, but very bad in others...

Warner