[LEAPSECS] Crunching Bulletin B numbers (POSIX time)

Joe Gwinn joegwinn at comcast.net
Sat Feb 19 11:08:08 EST 2011

At 3:59 PM +0000 2/19/11, Ian Batten wrote:

>On 19 Feb 2011, at 15:41, Gerard Ashton wrote:


>> On 2/19/2011 10:24 AM, Joe Gwinn wrote, in part:

>>> I have not been following the proposal in detail, but a key issue

>>>to the POSIX community is that their timescale must be

>>>implementable in a totally isolated machine, one having no GPS or

>>>internet access.


>>> There are other requirements as well. This was discussed at

>>>length on the Time Nuts reflector, until Tom kicked the thread

>>>over to Leap Secs.


>>> Joe Gwinn

>>> _______________________________________________

>>> LEAPSECS mailing list

>>> LEAPSECS at leapsecond.com

>>> http://six.pairlist.net/mailman/listinfo/leapsecs



>> I think a description of the totally isolated machine requirement

>>would have to be specified

>> to make progress. How long would the machine be expected to keep

>>good time, what is the acceptable

>> tolerance for time of day, and what technology would be used to keep time?


>With respect to leapseconds, a debate starting from a completely

>isolated computer is almost meaningless. Leapseconds represent a

>one part in ~5x10^7 divergence (0.02ppm) from TAI+initial offset.

>Even if you can find a free-running crystal oscillator of that

>precision given constant temperature, the environment in a piece of

>computer hardware varies in terms of temperature sufficient to throw

>that off by more than 0.02ppm, and (whisper it!) computers have been

>known to drop clock interrupts at a rate at least comparable to 1 in

>5x10^7. Show me a computer that can run its operating system clock,

>ie the return value from time(), such that it gains or loses

>significantly less than a second over the course of a year and we

>can talk about the semantics of Posix time relative to leap seconds

>in an isolated environment.


>Now it's possible that one scenario is a machine which is hooked to

>a Caesium-Rubidium clock, but has no means of obtaining DUT1 or leap

>second notifications, but nonetheless needs to track UT1 or UTC to

>high precision. I think that use-case needs to be articulated

>before it's used to hold this debate hostage, because I'm struggling

>to think what it is.

This is all true, but solves a problem that the POSIX Committee
neither understood nor cared about.

This was beaten to death on the Time Nuts reflector in the thread
"Leap seconds and POSIX" around January 2009.

Joe Gwinn

More information about the LEAPSECS mailing list