Re: [bess] WG Last Call for draft-ietf-bess-evpn-inter-subnet-forwarding-03

[Eric C Rosen <[e-mail]>]

I have a number of comments on this draft.  I've attached a copy of the 
draft with comments in-line; look for lines beginning with "****".

I don't think the document is ready to advance at the present time.

Issues:

- Most of this document is a discussion of various Data Center use 
cases, with an informal discussion of how EVPN procedures could be used 
to get IP datagrams into or out of a DC.  A little bit of the document 
is the specification of EVPN protocols and procedures that is specific 
to inter-subnet-forwarding.  However, these two parts are not clearly 
separated.  This makes it very hard to know which parts of the document 
are normative (i.e., suitable for Proposed Standard status) and which 
are just use case descriptions (as one might find in an Informational 
document).  This really needs to be fixed; I don't see how one would 
expect interoperable implementations to result from this document.

- These seems to be an architectual model hidden here, in which 'IRB 
interfaces' connect IP-VRFs to Bridge Tables (or something). However, 
there doesn't seem to be a clear description of this model.

- The use cases discussed in section 4 (Asymmetric Forwarding) are 
different than the use cases in section 5 (Symmetric Forwarding). 
However, I don't think there is an implication that certain use cases 
require asymmetric forward and certain require symmetric forwarding.  
I'm really confused about how to interpret sections 4 and 5.

- It is not always clear whether the discussion of use cases is or is 
not intended to be normative.

- The sections discussing the use cases contain a lot of text that is 
repeated verbatim (or almost verbatim) from other sections. This makes 
almost impossible to see what is done differently for the different use 
cases.  I think this repeated text needs to be refactored or removed.

- The discussion of routing packets between an "EVPN domain" (my term) 
and the "outside world" (Internet, IPVPN, other EVPN domain) does not 
provide much information on how one actually makes that happen 
correctly.  (The only thing that is really covered is routing between 
two subnets of the same tenant; everything else seems like just a 
placeholder for sections that were never actually written).

- Much of the terminology is not precisely defined, and normative 
references are not given to documents where the terminology is defined.

- No attempt is made to use a consistent set of terms.  This often 
leaves on wondering: "it says 'Broadcast Domain' here, it says 'subnet' 
there, it says 'MAC-VRF' in the other place, are these terms being used 
interchangeably, or is there some difference that needs to be attended 
to?".

- In a number of places, there seems to be a presupposition that an EVI 
contains one Broadcast Domain.  This is not true for all the variants of 
EVPN service.

In order for this document to advance, I think it needs the following:

- Decide whether it is a protocol spec or an applicability guide. If 
both, separate the normative from the descriptive part in a clear way.

- Clarify the architectural model.

- Eliminate the large sections of repeated text.

- Tighten up the terminology.

- Eliminate the sections that don't really say anything (e.g., the 
sections on routing between an IPVPN and an EVPN, the section on 
mobility).  Alternatively, provide content.

Having said all that, I would like to see this document go forward, but 
I don't think it is ready.

More comments in the attachment.