[KLUG Advocacy] Linux turning into Windows?

[Adam Williams]

> BTW Bruce your subject was negative from the beginning, so I guess the
> outcome of this thread might not be objective in the end.

Well, if the subject is negative then I disagree.   I think the concept
of a unified configuration system is excellent.

> > I think building a special key/value system for making a global registry is just
> > dumb.  LDAP & SLP with optional Kerberos are well established and tested.  But
> > this gee-i-think-ill-reinvent-the-wheel bent of Open Source seems unfatigable; 
> > they'll waste some man-weeks, give up, and accomplish consuming some sourceforge
> > disk space - much like the myriad Open Source groupware projects that reached
> > version 0.9.9 and fall over.
> Would you really run LDAP on so one to hold the system configuration like 
> IP-address of the network interface and other essential parameters ? Just 
> woundering.

Then I'd suspect that XML is the correct way to go.  The correct values
could be retrieved with something like xpath queries.  Lots of tools
already exist to create/manipulate such things.   One could even use
'high level' utilities like GNOME-Db's XML support to construct end-user
tools.  

And XML already has answers for multi-valued keys, binary (or just
non-UTF8) data, language tagged data, crypted values, etc...  Even if it
isn't part of the original design spec it seems inevitable that these
features would be requested, and added.  Much like MySQL was once the
fast&light database, and new versions will have constraints, triggers,
transactions, views, etc...  In one sense haven't they just recreated
PostgreSQL/Oracle/DB2/Informix that they originally decried because of
their resource consumption and complexity?  (Just as an example).

Not to mention a verification mechanism.  Apache could provide a DTD for
a valid configuration file, and any specific configuration could be
validated.  

Not that a daemon might not be neccesary (gconfd uses one) to manage
access and integrity, and an easy mechanism for processes to load and
unload bits of the store.

> > They'll discover that knowing nothing about content and offering 'simplicity'
> > will make this basically useless because of all those wierd edge conditions the
> > universe like to create.  This is Apple's NetInfo or OpenLDAP 1.x.x - both of
> > which the world has abandoned (or is in the process of abandoning) because of
> > all the problems they couldn't solve.  This has been hashed out ad infinitum in
> > various directory services forums.
> So what do you think where should the way lead for system configuration? 
> Something like Gconf (if that is file based at all), as network services 
> are not up yet when we need the data to configure the network, ....or 
> somewhere else?

Yes, I think something gconf-esque would be the best answer for that bit
of initial information.  [note that on a corporate network alot of this
is all provided by DHCP, so we are probably talking about stand-alone
and SOHO systems].   But once up and running managability via LDAP would
be a boon (again, tools are pre-exist), the master system could contain
referral objects to the individual systems (much like CIFS machine
account objects) and everything could be edited seemlessly with any
decent LDAPv3 tool.

And this would really be a whack against proprietary systems like AD and
NDS which provide this end-to-end managability.