I've been following this debate with great interest, and would like to
briefly give my views on the startup/shutdown and 'run level' stuff...

What I mostly miss is a clean way of starting and stopping parts of
the system.  Simple commands to start and stop, for instance, the IP
networking, kerberos services, the postgres dbms, appletalk, httpd and
so forth.  Further, I'd like to be able to combine these things into
groups of functions that I could give names to, and start or stop all
of such a group with just one command.

For instance, I could create a named collection 'dbms', and give a
command like, say, 'rc start dbms' from single user mode, which would
cause the network to be configured, kerberos to be started, and
finally postgres to be fired up.  This would probably be specified as
a set of commands to actually start postgres, along with a list of
dependencies on other packages ('ip', 'kerberos') that need to run
before postgres can be started.  As another example, I would love to
be able to say 'rc stop cap' and have all processes relating to the
Columbia Appletalk system be cleanly shut down in the right order.

Run levels would be a simple by-product of such a scheme, since it
would just be a matter of defining collections called '1', '2' etc.,
and using them the way System V normally does.

The hard part is a good way of handling interdependencies and making
sure that all required collections are started before the one you ask
for, as well as that all colletions that depend on one you are
stopping are stopped first.

-tih
-- 
Tom Ivar Helbekkmo
tih@Hamartun.Priv.NO