Re: [LEAPSECS] Rubber seconds & need for different timescales: was Stupid question from programmer: Why not just eliminate ALL UTC corrections ?

From: Ed Davies <ls_at_EDAVIES.NILDRAM.CO.UK>
Date: Fri, 05 Aug 2005 14:07:07 +0100

Poul-Henning Kamp wrote:
> ....
> Also, your UTS proposal is a total non-starter: Rubber seconds is not
> a usable solution.
> ....

Whether rubber seconds are usable or not depends on
what problem you intend them to be a solution to. If
you just want a monotonic timestamp stream to be able
to give ordered labels to events (e.g., file system
transactions) then they are pretty good.

Similarly, making sure that scheduled events don't
fall between the cracks of clock updates is a lot
easier if you steer the clock via rate changes rather
than make jumpy phase changes. E.g., the methods you
listed for creeping up on a time instant by binary
chop work well in these circumstances.

Of course, if you need to record intervals accurately
then rubber seconds are a disaster. You need to be
using a different clock. Similarly, if you need to
stay in phase with processes defined in SI seconds
then rubber seconds are not good to use.

There is no single way of labelling time instants which
is good for all applications. This is fundamentally
because the rotation of the Earth is variable and
unpredictable and because we don't have ubiquitous and
instanious access to some standard time. All we can
hope for is to have a set of timescales which are
related in a way which makes sense for the intended
users - including the important characteristic that
those users can understand the limitations of the
scales they use.

Ed Davies.
Received on Fri Aug 05 2005 - 06:08:19 PDT