[LEAPSECS] Easter and NTP

[Rob Seaman]

After nine years of this discussion, the most flabbergasted I've been  
is when I realized that asserting proper system engineering principles  
riled the mailing list more than my most extreme flights of rhetorical  
fancy.  So Pete has identified a system requirement, and Warner  
(rather too heavily excerpted below), has done a nice job at writing a  
problem statement.

A fundamental aspect of system engineering is to separate the  
definition of the problem from the specification of a satisfactory  
solution.  Each problem has many possible solutions.  Say again:  each  
problem has many possible solutions.

To constrain the solution parameter space, we have to discover the  
requirements implicit in the problem statement in front of us.  This  
process is rarely straightforward since the problem can only be  
understood in its full context (typically expressed as use cases).   
For timekeeping, this context is vast indeed.  There is a very strong  
benefit to writing down a set of requirements, however; the discussion  
can focus on the proper description of the problem before entertaining  
trade-offs among the many possible solutions.

Once a consensus is reached about the requirements for a project, a  
technically savvy crowd will have significantly less trouble agreeing  
on solutions that implement those requirements.

After those nine years - even following a very ad hoc process - the  
problem facing civil timekeeping is much better understood (by this  
group, anyway).  After nine years, however, all we still understand  
about the one-and-only notion ever proposed for WP7A deliberation is  
that they (whoever "they" are) think they've found a sly, zero effort,  
way to avoid the problem entirely.  This doesn't comprise a "solution"  
at all.

This is what my grandfather called a "lazy man's load" - piling up too  
many items into a single precarious heap, rather than making the two  
trips - with manageable loads - that are really required.

Civil timekeeping comprises two parts - two lists of requirements.   
There is the Earth orientation part that Pete is discussing, and there  
is the interval time part that Warner is referring to.  It isn't  
surprising that it is a challenge to find a single set of requirements  
that satisfies both.  It will ultimately prove much easier to do so -  
and to entertain solutions implementing those requirements - if we  
don't spurn the tools of system engineering.

Rob Seaman
NOAO
--

On Nov 9, 2008, at 2:15 AM, Peter Bunclark wrote:


> So a User requirement might be:

> The rythms of life, including the orbits of the earth and the moon,

> the rotation of the earth, and convenient sub-divisions of the  

> rotation

> down to nearly  the limit of human perception, shall be expressed in

> a single monotonic calendar.


On Nov 8, 2008, at 8:48 PM, M. Warner Losh wrote:


> Since most of the world

> runs on time_t, leapseconds necessarily introduce a time step (or a

> time error for the misguided folks that try to paper over the leap

> second by running time slower on leap day...).