Re: WG Review: NETCONF Data Modeling Language (netmod)

Hi -

> From: "Dave Crocker" <dhc2@dcrocker.net>
> To: "Eric Rescorla" <ekr@networkresonance.com>
> Cc: <iesg@ietf.org>; <ietf@ietf.org>
> Sent: Tuesday, April 22, 2008 10:03 PM
> Subject: Re: WG Review: NETCONF Data Modeling Language (netmod)
...
> Are they committed to doing the work?

The bulk of the work has been done (or close to it) for quite some time.
Ideally, it would have been done *before* the NETCONF protocol was
cast in concrete, but the NETCONF working group was not allowed
to define a modeling approach before finishing a protocol.
Without data models, the protocol is useless.  Consequently, there
are already numerous vendor-specific ways of handling modeling, and
even multiple approaches showing up some companies.  Not good.

> Do they have their own constituency?

All the major players in the devlopment of the NETCONF protocol,
as far as I know.

> Since the topic is not new, where have they been and why have they not 
> developed their own group consensus?

Previous requests for a BOF like the one held in Philadelphia were denied.
The various design teams have considerable common ground, and the
consensus of the folks who are actually doing work is in my opinion
pretty accurately reflected in the charter proposal.

> Rather than "perspectives" where are the technical concerns that Bert asked about?

As I see it, the key technical issues are these:

   1) Is there a need for a domain-specific language for network
       configuration management data modeling?   Experience
       in the field gives an unequivocal "yes".  GDMO, SMI, and CIM are
       a few examples of how folks have dealt with the shortcomings of
       the general-purpose tools available over the years. General-purpose modeling
       languages are both too much and too little, particularly with regard
       to issues of inter-version compatibility of models and interoperability.  Even if
       a language can represent an important semantic, there's still the question
       of whether that particular solution is compact and intuitive.  With some, to
       represent common constraints like uniqueness the designer had to resort
       to the equivalent of assembler language.

  2) Does it make sense to use an XML-based syntax for the "human-friendly"
       representation of data models?  For "industrial-strength" models the answer
       becomes more and more "no" as the model becomes larger and more
       semantically rich.   This is not a question of expressive power.  It's a question
       of providing a way to support development of *readable* standardized
       data models for NETCONF.

Forgive my impatience.  We went through this same debate twenty years ago
regarding ASN.1 and GDMO, and only slightly later in de-coupling SNMP SMI
from ASN.1  The acronyms may have changed, but the answers haven't.

Randy

_______________________________________________
IETF mailing list
IETF@ietf.org
https://www.ietf.org/mailman/listinfo/ietf