> I agree completely here. However, semantics in terms of site
> configuration have the nasty property of being quite difficult to relate
> between two discrete sites. The problems I address at my site, while
> generally similar when "copying stuff" or "executing things" might be
> difficult to relate to the stuff being copied and executed at your site.
Therein lies the rub. Denotation is inherently in the eye of the
beholder. You are not the same beholder as I. By the time I talking
about a useful level of abstraction for my site, I'm speaking what to
you is gibberish.
> Perhaps examples for building these meta layers are precisely what the
> original poster desires. For example, my site configuration deals with
Some stuff appears in the archives, but anyway...
I'm speaking expressly of a singular meta layer. You need
something to glue all the disparate layers together to produce answers
within the domain of interest.
The problem is actually sufficiently high level that we can't
speak concretely about anything that is seemingly relevant to
system administration. Well, we can, but we're going to keep coming
back to the same themes.
The meta layer I speak of is a domain of domains, or an object of
objects. For lack of qualifications, the tools are called C++,
smalltalk, etc, etc, etc.
We don't want that; we want something more restricted but still a
domain of domains layer, and how you can speak of static properties of
the domains system administrators deal in such that you can combine
the domains and see how the aggregate performs.
A lot of the system configuration problems are actually "how does
this thing perform?" spelled backwards. "I want this object to behave
like this. What must I say to get that behavior?". In other words,
you are asking a computer program to write a computer program for you
based on a specification. There is plenty of prior art here, but
we're climbing a tall tree.
A lot of configuration problems can be tractably specified short
of this point, but this is one of the major asymptotes you are
approaching.
_______________________________________________
lssconf-discuss mailing list
lssconf-discuss@inf.ed.ac.uk
http://lists.inf.ed.ac.uk/mailman/listinfo/lssconf-discuss
