Re: [Cbor] Benoit Claise's Block on charter-ietf-cbor-00-04: (with BLOCK and COMMENT)

[Alexey Melnikov <[e-mail]>]

Hi Benoit,

On Mon, Dec 19, 2016, at 12:48 PM, Benoit Claise wrote:
> Benoit Claise has entered the following ballot position for
> charter-ietf-cbor-00-04: Block
> 
> When responding, please keep the subject line intact and reply to all
> email addresses included in the To and CC lines. (Feel free to cut this
> introductory paragraph, however.)
> 
> 
> 
> The document, along with other ballot positions, can be found here:
> https://datatracker.ietf.org/doc/charter-ietf-cbor/
> 
> 
> 
> ----------------------------------------------------------------------
> BLOCK:
> ----------------------------------------------------------------------
> 
> Sorry for this late BLOCK.
> I had a very quick call with Alexey before the last IESG telechat: I want
> to understand if I missed anything.
> I filed a quick "NO RECORD" COMMENT.
> Then, we discussed some more during the telechat itself.
> And now, I finally had the time to think some more about this.
> 
> My BLOCK is about this charter paragraph:
> 
>     Similar to the way ABNF (RFC 5234/7405) can be used to describe the
> set of valid messages in a text representation, it would be useful if
> protocol specifications could use a description format for the data in
> CBOR-encoded messages. The CBOR data definition language (CDDL, based on
> draft-greevenbosch-appsawg-cbor-cddl) is a proposal for such a
> description technique that has already been used in CORE, ANIMA, CDNI,
> and efforts outside the IETF. The CBOR WG will complete the development
> of this specification by creating an Informational or Standards Track
> RFC.
> 
> 
> In OPS, we need automation. And automation will come from data models as,
> from data models, we're able to generate APIs.
> In the world of data modeling-driven management, we have:
>     YANG as a data modeling language, with ABNF specifications
>     YANG data modules, written with the YANG data modeling language
>     different encodings, such as XML, JSON, or CBOR
> (draft-ietf-core-yang-cbor-03)
>     protocols such as NETCONF or RESTCONF for
> configuration/monitoring/capabilities discovery
>     note: working on pub/sub protocol (aka telemetry) for events
> 
> See the first picture at
> http://www.claise.be/2016/12/yang-opensource-tools-for-data-modeling-driven-management/
> Btw, I should add cbor.
> 
> Now, in this proposed WG, you want to define a new data modeling
> language, "The CBOR data definition language"
> When I ask the question: So the structure of what could be accessed on a
> managed device?, you answer:
> 
>     No. While CDDL could be used to describe the structure of data at
> rest (a data store), that is not the objective. CDDL is used to define
> the structure of data in flight, e.g. a protocol message going from a
> node to a different node. (Using a term popular in semantic
> interoperability work in the health care domain, CDDL is about
> "structural interoperability” — it can tell you that there is supposed to
> be a data item “cheese-firmness” in the message and out of what set of
> values it needs to come, but it cannot tell you what the specific values
> mean in the real world or what cheese firmness is in the first place on a
> semantic level.) 
> 
> 
> But what about the semantic definition (which is in YANG modules) of this
> information? This is key for management.
> I guess that the next item you'll want after this milestone is exactly
> data models and semantic, right?
> 
> There are many schemas for IoT and I'm not trying to impose YANG in the
> IoT world but I want to understand why we need duplication.
> Note that there was an IAB-organized workshop on IoT data modeling in the
> past (https://www.iab.org/activities/workshops/iotsi/)
> However, it seems to me that your effort is exactly the reverse of data
> modeling driven management? You have an encoding, and then you want a new
> schema language

Yes, the same way there are several schema languages for XML (for
example).

> Next, you'll need a mechanism to discover what is available on the
> managed devices, a mechanism to know the device capabilities, a mechanism
> for pub/sub, ...
> And you will redo the full OPS stack, for IoT (hence duplication). And,
> obviously, in the end, we will need a mapping between the two data
> modeling languages: YANG and CDDL.

I am not convinced that the mapping would be needed in all cases. See
below.

> What is specific here?  I wanted to write: what's specific to IoT here,
> but I don't even see IoT in the charter. There is just a kind of IoT
> reference in RFC7049 abstract.
> Why do we need this duplication within the IETF?

I don't think saying that CBOR is only for IoT is helpful. There is some
talk about using it for PKIX-like things. I can also see it being used
directly in protocols. There is no YANG there, because there are no
devices to manage, just a data structure. I don't think YANG is going to
be helpful there.

I am Ok with having some text in CDDL saying that if you want to do
modeling-driven device management, CDDL is not the right tool. But as I
said above I see other uses for CBOR/CDDL, which should be allowed.

> Why don't draft-ietf-core-yang-cbor and draft-vanderstok-core-comi work?
> Those are completely inline with data modeling-driven management and this
> charter seems to contradict this effort?
> What do I miss?