Re: [netmod] YANG Mount = Alias Mount + Peer Mount (was RE: Motivations for Structuring Models)

Hi,

I think the NETCONF WG really needs to think about the architecture
it is creating, and the role of proxy servers in network management.

When we were designing NETCONF (pre-4741) we were told by
knowledgeable operators like Randy Bush that NETCONF MUST NOT
have any Proxy-based management, because it added too much complexity.

   To date systems built upon YANG models have been missing two
   capabilities:

   1.  YANG Datastore Mount: Datastores have not been able to proxy
       objects located elsewhere on the same device, or upon a different
       device.  This puts additional burden upon applications which then
       need to find and access multiple locations and which may be on
       remote systems.

   2.  Eventual Consistency: YANG Datastore implementations have
       typically assumed ACID [1] transaction models.  There is nothing
       inherent in YANG itself which demands ACID transactional
       guarantees.  YANG models can also expose information which might
       be in the process of undergoing convergence.  Since IP networking
       has been designed with convergence in mind, this is a useful
       capability since some types of applications must participate
       where there is dynamically changing state.

The NETCONF WG should decide first "we need proxy servers" before debating
the best way to do that.

We have been told there are systems that converge very slowly (although
nobody seems
to know if that means 5 minutes of 5 hours).  All the "get-actual" type of
proposals
seem to address this issue.  Not sure how YANG Mount addresses the issue,
and how it is related to the other solution proposals.

Andy

On Tue, Sep 15, 2015 at 8:21 AM, Eric Voit (evoit) <evoit@cisco.com> wrote:

> There was a recent thread on structuring YANG models so that application
> developers might be able to reference alternative local hierarchies/tree
> structures for certain objects.  This thread motivated Alex, Sander, and I
> to rework the YANG Mount requirements draft.  v03 is posted at:
> http://datatracker.ietf.org/doc/draft-voit-netmod-peer-mount-requirements/
>
> This draft has been retitled to "Requirements for mounting of local and
> remote YANG subtrees".  This retitling was done because we have separated
> the thinking on what it takes to Mount objects from remote devices (Peer
> Mount) from what it takes to Mount within the same device (Alias Mount).
>
> We would be interested in your thoughts.
>
> Eric
>
> -----Original Message-----
> From: Ladislav Lhotka, August 31, 2015 11:05 AM
>
> Randy Presuhn <randy_presuhn@mindspring.com> writes:
>
> > Hi -
> >
> > It is with no little amusement that I watch this thread struggling
> > with questions that were solved fairly neatly a quarter century ago in
> > GDMO/CMIP-land.  I'm *not* suggesting we go back there, but would like
> > to offer an observation about modeling that might help.
> >
> > The organization of instance data in SNMP is a direct mirror of the
> > "object" definitions.  Simple at first, but quickly becoming baroque
> > as various minds of "multiplexing" are added to compensate for post
> > hoc deficiencies in the index structures.
> >
> > Life is such that once a resource has been modeled, it will be
> > used/re-used/embedded in systems in ways in which its designers
> > couldn't be expected to imagine.  A consequence of this is that if
> > instance naming is completely locked down when the management
> > interface for a resource is first defined (as it is in SNMP) then all
> > sorts of peculiar hacks will be needed to deal with, for example,
> > virtual routers.  Unfortunately, an SNMP/SMI-like mindset is so
> > pervasive that folks seem to overlook that there are other ways to
> > deal with this situation.
> >
> > What GDMO did was to use a separate "NAME BINDING" construct to
> > specify contexts in which instances might show up, allowing instances
> > to be put in places that weren't even imagined when the original class
> > definition was written.  Name bindings could be standardized, or be
> > vendor or even product-specific, allowing the simplicity or complexity
> > of a given system's instance tree to reflect the actual simplicity or
> > complexity of that system, rather than requiring all systems to be
> > structured for the worst case.
>
> How could this be expressed in YANG terms? (I tried to figure it out
> myself but I unfortunately couldn't make any sense of sec. 8.6 in CCITT
> Recommendation X.722).
>
> Thanks, Lada
>
> >
> > Yes, separating the specification of instance naming in large part
> > from class definition does have implications for how one does access
> > control, and how clients figure out how to ask a server to create
> > something, but it's not a huge deal - it's just not like VACM, and a
> > whole slew of hacky solutions and "wierd plumbing adapters" (to borrow
> > from Jeff Case) just go away.  Strangely, it makes the job of the
> > initial modeler and of the eventual user much easier.
> >
> > Randy
> >
> > _______________________________________________
> > netmod mailing list
> > netmod@ietf.org
> > https://www.ietf.org/mailman/listinfo/netmod
>
> --
> Ladislav Lhotka, CZ.NIC Labs
> PGP Key ID: E74E8C0C
>
> _______________________________________________
> netmod mailing list
> netmod@ietf.org
> https://www.ietf.org/mailman/listinfo/netmod
>