Re: [i2rs] 'network type' placement and RFC8345

[tom petch <[e-mail]>]

From: Qin Wu <[e-mail]>
Sent: 24 September 2020 13:17

-----邮件原件-----
发件人: tom petch [mailto:[e-mail]]
发送时间: 2020年9月24日 16:45

From: Qin Wu <[e-mail]>
Sent: 24 September 2020 03:22

Hi, Tom:
Layer2, Layer 3, TE are all base modules which other modules can extend from.
I am not sure we have Layer 1 base module, WSON and flexi-grid, if my understanding is correct, are TE technology specific and WSON and flexi-grid module can be seem as extension to TE module or a module derived from TE module Therefore we could follow OSPF example defined in the L3 topology module or L3 TE module defined in draft-ietf-teas-yang-l3-te-topo-08.

<tp>
Qin

I think that you are missing the point.
RFC8345 sets up a registry, a namespace, and gives rules about how it should be used.  The use of a YANG presence container is clearly specified.
What I find unclear in RFC8345 is the intent about tree structure it describes, what is the criterion for placing e.g. network type 802.3 alongside or below one of the existing network type.
How the different network type YANG modules relate with respect to YANG import and augment is a different and irrelevant question IMHO.
My thinking is that CCAMP have got it wrong and that their network types should be alongside existing types, that there is nothing in RFC8345 to suggest that they should be subordinate to anything else.
Since they are presence containers I see nothing in YANG that would make a tree structure anything other than more complicated with a longer path to reference them bringing no benefit.

Tom Petch

"
   module: ietf-l3-te-topology
     augment /nw:networks/nw:network/nw:network-types
               /l3t:l3-unicast-topology:
       +--rw l3-te!
"
"
   module example-ospf-topology
   augment "/nw:networks/nw:network/nw:network-types/l3t:l3-unicast-topology"
       +--rw ospf!
"
I might be wrong if a generic layer 1 can be defined without adding dependency to TE technology. But at least layer 1 type or layer 0 type are common building block that can be reused.

In addition, base model, in my opinion doesn't need to limit to layer 1, layer 2, layer 3, service layer, TE layer this angle, we may classify network topology from other angle, e.g., classify network topology into UNI topology and NNI topology, One relevant model is UNI topology model that is proposed in the opsawg
https://tools.ietf.org/html/draft-ogondio-opsawg-uni-topology-01
such models are also base model which other modules can derive from.

For network type, if we can define it as identity, it may be another design option.
But comparing with presence container design, I think the only difference is one is explicit way, the other is implicit way.

<tp>
Qin
My starting point is RFC8345 which for me creates this 'registry' of network type and lays down the ground rules.  It says that a presence container must be defined.  Why not identity, as with routing protocols, I do not know.  It suggests a tree structure and is generally keen on a layered network as in layer 1, layer 2, layer 3, application-related layers and so on which is fine until you get to sub-IP.

But, network layering has nothing to do with YANG modules, with imports, cross-references, dependencies and so on so the fact that the wson module augments te-topo seems an irrelevance where the tree structure of layers is concerned.  There are lots of augments to te-topo and they could be any layer.

[Qin]: we can have ietf-l3-te-topology module augment to layer 3, but we don't have l3 specific module augment to te module, see draft-ietf-teas-yang-l3-te-topo-08.
I will not see ietf-te-topology-packet as l3 specific module augment to te module. But may be a little bit fuzzy.
For ccamp modules which are classified into layer 0,layer1 and married with te technology, I feel nature they augment from te module, but not other layer module. That's my impression. But I might be wrong.

RFC8345 says a lot but I do not understand what it is saying when it comes to adding a network type beyond what it says about presence container.  What is the point of the tree structure therein?
[Qin]: The question is whether there are any layer0, layer 2 modules without marriage with TE technology?

 Is it something we want or need to embody in YANG so that we can do something fancy as we can elsewhere with base identity and derivations therefrom as with routing protocols?

I do not know so my inclination is to say that the structure should be flat until we have a good reason otherwise lest we find ourselves tied up in knots at a later data.  I think it would be quite wrong to build a tree structure of network type based on YANG import which is what I suspect CCAMP are doing.

Tom Petch

-Qin
-----邮件原件-----
发件人: i2rs [mailto:[e-mail]] 代表 tom petch
发送时间: 2020年9月23日 17:16
收件人: Sue Hares <[e-mail]>
抄送: '[e-mail]' <[e-mail]>
主题: [i2rs] 'network type' placement and RFC8345

RFC8345 requires that a new network type be given a presence container and suggests a tree structure with layer 1, layer 2, layer 3 and service as top level nodes with OSPF as an example of a node subordinate to layer 3.  te-topology , RFC8795, places its presence container at the top level alongside these four.
Question; where should a network type such as WSON or flexi-grid be placed?  wson-yang, in IETF Last Call, places it under te-topology which is possible but it seems to me more like a layer 1 or layer 0. But then network types do not seem to form a tree, rather a mesh so a tree structure seems wrong.  And wherever layer 1 is defined it is not in a module imported by wson-yang although it might be added to layer0-types (!) which wson-yang does import. I would see it as wrong to define layer 1 in wson forcing others to import wson.

Thoughts?

I have posted this to Lou and TEAS but as it is a question that cuts across multiple WG I suspect that I will get multiple contradictory answers or none:-)

Tom Petch
_______________________________________________
i2rs mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/i2rs