Review comments on draft-ietf-rtgwg-ni-model-01

[Robert Wilton <[e-mail]>]

Hi,

I've reviewed draft-ietf-rtgwg-ni-model-01.  Generally I approve and 
agree with the structure, but I have a few main questions/comments 
followed by some minor comments:


Main comments/questions:

1) Has it been considered whether, and how, the IEEE 802.1 bridge YANG 
model (802.1Qcp) should interact with this network instances model?  It 
would certainly be nice if an 802.1 bridge could possibly sit under a 
network instance rather than either always being top level, or under a 
completely separate hierarchy.

2) For the binding between interfaces and network instances the model is 
using a string rather than a leafref to a network-instance name leaf.  
Naively it would seem that a leafref would be better than a string since 
it would ensure that an interface cannot be bound to a network instance 
that doesn't exist.

3) I presume that there is a notional default network instance that all 
interfaces/sub-interfaces belong to if they haven't been assigned to a 
specific network instance.  Assuming that this is the case it might be 
useful if the document mentioned this.

4) The YANG model in chapter 6 is somewhat IP centric.  It might be 
cleaner if there was a single top level network-instances model that 
covered by L2 and L3, and for the IP augmentations to be in a separate 
YANG model that augments the layer independent network instances model.

5) The model allows different address families on the same interface to 
be put in separate network instances, but does it allow for an interface 
to be put in one network instance, but a particular address family on 
the same interface to be put in a different network-instance?

6) Considering sub-interfaces, I assume that putting a parent physical 
(or LAG) interface into a network-instance doesn't also implicitly place 
any child sub-interfaces in the same network instance.  I.e. when 
considering network-instances each sub-interface can be regarded as a 
discrete interface object independent of its parent?

7) The OpenConfig network-instances model 
[https://github.com/openconfig/public/tree/master/release/models/network-instance] 
allows for a combined L2/L3 network-instance (type = L2L3).  Is the 
intent that the IETF network instance model allows such a construct?  Or 
would this always be modeled using two separate network instances?  
Should the document discuss this point?


Minor comments:

8) Section 1.1. top open issues:
This could now hopefully be updated to reference 
draft-nmdsdt-netmod-revised-datastores-00 (or the WG version if it 
succeeds in getting adopted).

9) I note that the binding from interfaces to network instances is 
specified on the interface, rather than as a list of interfaces as part 
of the network instance.  I agree with the current approach because the 
structure of the model trivially enforces that an 
interface[+address-family] can only be bound to a single network instance.

10) The YANG tree output in Section 2: Overview, page 3:
- The diagram shows bind-network-instance-name under ethernet, but I 
think that it should be shown directly under "interface".  (As an aside, 
I'm not convinced whether lldp should be under ethernet since it should 
run over other physical layers.)

Regards,
Rob