[LEAPSECS] presentations from AAS Future of Time sessions

[Magnus Danielson]

On 14/01/14 16:37, Warner Losh wrote:

>

> On Jan 14, 2014, at 8:05 AM, Steve Allen wrote:

>> In 1980 November the CCITT accepted UTC "as the time scale for all

>> other telecommunications activities".  In 2007 the BIPM contributed

>> document 7A/51-E to the ITU-R WP7A meeting regarding Question ITU-R

>> 236/7 saying please don't use TAI, we might even suppress it.  Then in

>> 2010 November ITU-T SG 15 recommended the use of PTP (IEEE 1588) in

>> ITU-T Recommendation G.8265.1, an operational time scale based on TAI.

>>

>> These international agencies with multi-letter-acronym names are

>> still not listening to each other about the nitty gritty details.

>

> To be fair, TAI in IEEE 1588 isn't the same TAI as the BIPM's (or more accurately TAI(BIPM)), but rather it is a constructed time scale where every second is enumerated consecutively with no second being second class (leap second) like in other time counting things. The "TAI" specified is really just a count of seconds since the epoch, inclusive of leap seconds, with an epoch chosen to line up with time_t + positive_leaps - negative_leaps. There's some ambiguity in the spec about how to encode times before 1972, but since it is a time exchange protocol that ambiguity doesn't matter. And IEEE 1588 is all about measuring the offset between clocks (in phase and frequency), but doesn't say anything about how to correct that, let alone how to present time to the user. They really mean 'number of PPS seconds since 1970, with some clever math to avoid the mess during 1970 and 1971 with rubber seconds and millisecond leaps'

>

> It is very nearly the same time scale that I defined for an internal without leaps timescale at a prior job. I defined it as 'number of SI PPS pulses since 0:00:00 Jan 01, 1972 UTC + 63072000. If you do the math, this is the same as 'number of seconds since 1970, counting leap seconds' but removes the rubber seconds, and millisecond leaps from before 1972 which make that imprecise statement .... I could also have called the time scale "TAI - 10s" and informally that would be right, but it wouldn't track to the BIPM's TAI time scale.

>

> So it is unfortunate they chose the name 'TAI' to mean this, but even if the BIPM suppressed TAI, the enumeration that aligns to TAI would remain.


TAI or shifted versions of TAI using some epoch offset and some set of 
gears is used by many systems. Any such TAI-like scale isn't *THE* TAI, 
but aligns up like TAI. With a bit of not too complicated handwaving the 
relation to TAI can be achieved and thus basis for traceability produced 
if you need that. It's just that use of TAI for this purpose has been 
actively discouraged, despite the fact that it is needed and is being in 
use.

The so called proliferation of UTC is really a proliferation of TAI. Had 
they actually promoted the duality of time-scales that UTC and TAI 
represents, the situation would possibly have been different.

They have been trying for forge two incompatible concepts into one.

Cheers,
Magnus