[LEAPSECS] Crunching Bulletin B numbers (POSIX time)
Warner Losh
imp at bsdimp.com
Tue Feb 22 16:31:53 EST 2011
On 02/19/2011 15:13, Tim Shepard wrote:
>
>> this is why leap seconds announced ten years in advanced
>> are important: they allow for a stand-alone machine, albeit
>> one that only needs to have it's software upgraded once in
>> ten years.
>
> A stand alone machine is going to have its clock drift by more than a
> few seconds over 10 years. Leap seconds are in the noise compared
> to errors due to accumulated drift of standalone machines.
>
> If it has a way of receiving a time signal to stay synchronized, then
> it ought to have a way of receiving info about the leap seconds (if
> they even matter).
Having a known list of leap seconds let one recover TAI time from a cold
GPS receiver in a few seconds to a minute, rather than waiting ~20
minutes for the almanac to download. Many systems have their internal
clocks in TAI, but also have requirements to report time in UTC to
users. Such a table would ensure these systems could come back online
quickly without having to pay the 20 minute penalty. While the system
may have already been down a little while when the swap is made, an
extra 20 minutes can kill reliability targets for the system.
As someone that's had to have their real-time system in a time-scale
that doesn't have leap seconds, but also report timestamped events in
UTC to the user, the 10-year horizon solves many problems with leap seconds.
As to Posix, I think the standalone system is in the noise compared to
the amount of code that actively depends on leap seconds not existing.
It might make a good talking point, but if you eliminate that
requirement, you still have the dusty deck problem to cope with.
Warner
More information about the LEAPSECS
mailing list