On Thu, Sep 10, 2015 at 11:43:29AM +0200, Benoit Claise wrote:
> Jürgen,
> >On Wed, Sep 09, 2015 at 08:38:35PM -0400, Nadeau Thomas wrote:
> >>    I wanted to set things up for the interim meeting tomorrow. To frame 
> >>    the meeting, we want to achieve two main goals:
> >>
> >>    1) close on requirements around a requirement to define a structure 
> >>    for IETF models and the requirements around ops state/models
> >>
> >I am confused. Benoit pointed to section 4 which has essentially
> >requirements concerning the support of asynchronous systems. You seem
> >to be talking also about requirements that go beyond that. It would be
> >good to know well ahead of time what is exactly on the agenda.
> I also mentioned: "or around the requirement that the YANG models need 
> some sort of hierarchy (draft-openconfig-netmod-model-structure), like 
> for the routing models, ..."
> 
> If the group acknowledges that there is a requirement to structure all 
> YANG models, basically extending what's done for the routing models, 
> that would already be an achievement.
> I would like to verify whether my understanding is correct: there is a 
> willingness to structure the YANG models at least per technology 
> (routing, OAM, for example), and we're debating whether or not "/device" 
> is appropriate. This "/device" is a minor point IMO.

I tried to explain this in one of my previous posts. There is a real
challenge ahead to design 'good' models, namely models that are
extensible and maintainable and that remain interoperable over a long
lifetime. This requires to identify and model key core components or
abstractions and getting them 'right' is hard and requires lots of
detailed reviews, discussions, revisions. To me, the discussion where
to root models is somewhat missing the real challenge in front of us.

That said, I believe the previous discussions that led NETMOD to use
the flat roots we have used in various RFCs that define core models
have not been proven broken in the sense that the roots are flawed up
to the point that we should declare failure and throw away RFCs and
running code. And I like to note that models such as ietf-interfaces
or ietf-routing actually provide structure (e.g., interface specific
models are expected to augment the core interface model) but they go
way beyond defining a simple root path. It is these kind of models
that we need. We also need to understand the tension between cohesion
and coupling of data models. Having a single model that every other
model directly or indirectly has to augment provides a coupling that
may prove to be a major headache in 10-15 years from now.

Back to the subject of the thread, the discussion around the support
for asynchronous systems and the discussion around the value of a
common nested root structure for models are in my view two separate
things and we should not mix them.

/js

-- 
Juergen Schoenwaelder           Jacobs University Bremen gGmbH
Phone: +49 421 200 3587         Campus Ring 1 | 28759 Bremen | Germany
Fax:   +49 421 200 3103         <http://www.jacobs-university.de/>

_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod

Reply via email to