[LEAPSECS] The relation between Easter and leap econds.

Rob Seaman seaman at noao.edu
Mon Nov 10 09:01:14 EST 2008


This is an excellent example of how focusing on the requirements
(rather than notional solutions) helps to clarify the discussion. For
one thing, the issue becomes one of describing the problem before
trying to figure out what to do about. For another, the discussion is
abstracted from entrenched positions. For a third, we start to reveal
underlying points of agreement (eg., disdain for politicized issues),
not just differences. And for a fourth, the process finds a natural
pace. If there are important scheduling drivers, what are they? In
short, what is the hurry?

Regarding the specific points here:

1) Yes, indeed! Requirements change. This is all the more reason to
focus on characterizing them now, and as they evolve. How else to
figure out what's going on?

2) Describing the process is separate (although entangled) from
describing the problem. I could wish some other process were being
followed, most obviously that whatever standards body this drama is
played out in, the deliberations should be open.

3) Virtually all system engineering is for the benefit of users who
are non-experts. This does not distinguish timekeeping in any special
way. What users want and what they need are often very different.
They may not care (or think they care - or even be aware) about time,
but that doesn't mean timekeeping doesn't enter the user requirements
in numerous ways.

4) A large project will have a long list of requirements. Timekeeping
is deceptively simple. Stepping aside from solar time for a moment,
just think of the very complex system of systems that generates and
delivers interval time, whether "atomic" or GPS. Ensembles of very
complex clocks. Multiple interlocking strata of NTP servers. Host OS
support. (Bad solutions are likely to be even more complex than good
ones :-) Etc and so forth.

5) Elected government officials are already involved. They changed
DST last year (and not just in the U.S.) The ITU is an assemblage of
government representatives, only by happenstance are they scientists.

6) To focus on the requirement at hand for a moment: Pete Bunclark's
comment about "a single monotonic calendar" points out that we already
have this situation. A mean solar clock is a subdivision of the
calendar. Modifying UTC to eliminate leap seconds would violate this
requirement. It is better to state and debate - and yes, perhaps
reject - the requirement, than it is to leave the requirement hidden
and violate it for some imagined short term political expediency.

6a) Why? Because if the requirements aren't stated, then engineers
and decision-makers creating contingent systems (and there are many,
many systems contingent on timekeeping) can draw diverse - and
diversely wrong - conclusions about the behavior of time. This is
precisely the situation decried now with leap seconds.

6b) It is not sufficient to describe a solution. If the problem that
a system or standard is intended to solve is left poorly described,
stakeholders won't comprehend the context to apply it appropriately
across all (or perhaps any) situations. Again - see leap seconds. A
hasty elimination of leap seconds will leave them as the proverbial
dog that didn't bark - not present, but not absent either.

7) Requirements are not demands from stakeholders (although
stakeholders make lots of - often incoherent - demands :-) Rather,
proper requirements form a coherent and self-consistent set of
descriptors of the problem and project at hand. Think of them as an
orthogonal set of basis vectors in a multidimensional parameter space
that captures the essence of the problem and supplies constraints on
the acceptable solutions.

Rob Seaman
NOAO
---

Poul-Henning Kamp wrote:


> Tony Finch writes:

>

>> My point is that user requirements change. The relation between

>> natural

>> phenomena and our consensus idea of time depends on how accurately

>> and

>> quickly we can measure those phenomena, and which phenomena we care

>> about

>> tracking.

>

> Couldn't the opposite argument be made: The majority of users don't

> care

> a hoot about time (ref: flashing 12:00 everywhere) and therefore the

> scientists and engineers who have to care, are free to implement

> any working solution ?

>

> After all, with the cavalier attitude politicians have to time, (ref

> timezone changes with short notice, century old laws etc), I can't

> see any good coming out of involving them.

>

> In other words: WP7A may not be the best place, but it is probably

> as good as any.



More information about the LEAPSECS mailing list