[Bier] BIER/IPv6 Requirements and Solutions

[Alvaro Retana <[e-mail]>]

Hello all!

I am directing this message to the authors of the BIER in IPv6-related
work.  I am also copying the WG because this work needs more
engagement[*], making sure I don't miss anyone.

  [*] More engagement in the form of substantive discussions from non-
  authors, and diversity of affiliations.


Some of you raised concerns to the IESG about the speed at which the
IPv6-related work has been moving, and the ability to discuss it
during WG meetings.  In response, I am providing my opinion of the
current state and, after consultation with the Chairs, reinforcing the
required steps to move any work forward.

Note that my intent is not to insert myself in the discussions, but to
get them back on track.  All decisions should be made by WG consensus
as interpreted by the Chairs [rfc2418].


I have done a preliminary/high-level review of the requirements
document (see details below); I don't think it currently is in a state
to provide adequate material for WG discussion.  Besides specific
requirements, the text should include a clear justification for the WG
to reach a consensus.  The current version doesn't properly cover
these aspects.

The document points to the WG Charter in a couple of places as
justification for the work.  Recommendation: Focus on the technical
aspects and not whether something is in the Charter or not!  If there
is WG consensus, a Charter can be amended to include new work.  On the
other hand, while a Charter is used to provide scope and focus, it
must not be used as justification for doing work in the absence of
discussion and consensus.


The critical step towards adopting a solution (or solutions) is
engaged discussion of the requirements.  After looking at the document
(comments below), I don't think it provides a good base for analysis.
The expectation is for the requirements to be clear, specific,
measurable, and to justify their applicability in the solution.

In consultation with the Chairs, we expect a revised requirements
document *before* IETF 109.  As I mentioned above, more engagement is
needed in its development.  Ideally, the text will be discussed on the
list as it evolves.  If a live discussion is required, the Chairs will
organize one (an interim meeting, if there is enough time before IETF
109).

Greg is the Chair responsible for this process.  [Side note: Tony has
decided to remove himself as co-author of one of the proposals.  His
comments will be considered as coming from a WG participant.]


If anyone has concerns about this message, please contact me directly.
Again, I won't participate in the WG discussions, but (if needed) may
provide additional feedback related to my review of
draft-ietf-bier-ipv6-requirements-06 (below).


Thanks!!


Alvaro.





Review of draft-ietf-bier-ipv6-requirements-06

This is an informal review.  While I expect this document to serve as
a discussion point for the WG, and for it to *not* be published as an
RFC, I have strong concerns about its contents.

Note that the intent of providing a review at this point is to aid in
progressing the IPv6-related discussion.  Any consensus on the
specific requirements should come from the WG, guided by the Chairs.

Consider all my comments as my personal opinion.


===

In general, the requirements listed are either too vague or not well
justified.  The objective "to help the BIER WG evaluate the BIER v6
requirements" cannot be met unless there is clarity that can lead to
engaged discussion.


The document starts by explaining the problem space (from the Introduction):

   As clarified in the working-group, "BIER natively in IPv6" means BIER not
   encapsulated in MPLS or Ethernet.  This may include native IPv6
   encapsulation and generic IPv6 tunnelling.


I didn't look at the archive, but I'm sure the WG can do a better job!!

This description is vague (""...natively in IPv6" means BIER not
encapsulated in MPLS or Ethernet"), recursive (native "may include
native") and confusing (native "may include native...and...tunnelling
[sic]").

Note that the Problem Statement itself ("to transport BUM packets,
with BIER headers, in an IPv6 environment.") is straightforward -- the
expectation of what is required as a solution is not.


Two conceptual solution models are presented.  The description of the
"Transport-Independent Model" is not in line with the layering model
from rfc8279.  In the best case, the terminology doesn't match; but I
think that the description might introduce new concepts.  Note also
that the statement that "BIER-MPLS could use this approach directly
since BIER-MPLS is based on MPLS" seems to contradict the definition
of "natively in IPv6".

The description of the "Native IPv6 Model" gets a slightly different
treatment; for example, it includes benefits.  Note that this model
also mentions "a trusted IPv6-based domain" while comparing the model
to SRv6, *and* talking about a "wider inter-AS scope".  These seem to
be contradictions as a trusted domain is typically aligned to a single
administrative entity [rfc8402].  Again, the terminology may not be in
alignment...or more discussion may be needed on the scope.



On to the requirements.  As far as requirement levels go, I recommend
that you use three levels instead of two: required, recommended, and
optional.  The level should be justified depending on the potential
deployment or solution model.


Some of the mandatory requirements listed are obvious and probably
don't need even to be mentioned.  That is the case for:

352	4.1.2.  Support BIER architecture
361	4.1.3.  Conform to existing IPv6 Spec
368	4.1.4.  Support deployment with Non-BFR routers


Some of the requirements are not clear -- they need a better
description and clear justification.

345	4.1.1.  L2 Agnostic

347	   The solution must be agnostic to the underlying L2 data link type.
348	   The solution needs to support P2P ethernet links as well as shared
349	   media ethernet links without requiring the LAN switch to perform BIER
350	   snooping.

Agnostic to the L2 data link, but also needs to support specific
topologies.  That is a contradiction.  Also, where is "BIER snooping"
defined?


376	4.1.5.  Support inter-AS multicast deployment

378	   Inter-AS multicast support is needed for ease of provisioning the
379	   P2MP transport service to enterprises.  This could greatly increase
380	   the scalability of BIER, as it is usually considered to be suitable
381	   only for small intra-AS scenarios.

rfc8279 talks about a single BIER domain.  There is no discussion
about BIER domains in the context of inter-AS operation in this draft
or draft-geng-bier-ipv6-inter-domain.  Again, in the best case, we
have a terminology mismatch. Still, it is not clear if the text
proposes new functionality -- and using
draft-geng-bier-ipv6-inter-domain as the base for a mandatory
requirement doesn't seem correct without WG consensus.


383	4.1.6.  Support Simple Encapsulation

385	   The solution must avoid requiring different encapsulation types.  A
386	   solution needs to do careful trade-off analysis and select one
387	   encapsulation as its proposal for best coverage of various scenarios.

Based on the description of "natively in IPv6", it seems like anything
that is "not encapsulated in MPLS or Ethernet" should be ok.  What is
the justification behind one encapsulation?  It sounds like this
requirement opens the door to multiple solutions, each with different
encapsulations -- doesn't that contradict a mandatory requirement of
one?


Finally, we find the security requirement:

389	4.1.7.  Support Deployment Security

391	   The proposed solution must include careful security considerations,
392	   including all that is already considered in BIER architecture RFC8279
393	   and RFC8296, and other security concerns that may raise due to the
394	   addition of IPv6.

I want to highlight the obvious nature of "must include careful
security considerations", which is a documentation issue -- there are
no specific security requirements on the solution itself?  No
particular mention of any security aspect.  What is the technical
need?  :-(


The optional requirements don't fare much better.  In general, please
clarify and justify them.

Some of the optional requirements seem to be associated with
unexplained scenarios.  For example, the support for MVPN or
IPSEC...and the specific need to forward "in hardware fast path".

Others seem to make improper assumptions (for example, as far as I
know, there is no existing standard OAM method) or suggest potential
enhancements ("enhance the capability of BIER leveraging the BIER-MTU"
 ??).


In summary, the list of requirements presented is not clear or
adequately justified.  It is not conducive to a WG discussion about
the requirements themselves...much less about potential solutions.