Re: [Bier] draft-ietf-bier-ipv6-requirements-09

[Greg Shepherd <[e-mail]>]

GS - inline:

On Wed, Nov 18, 2020 at 7:14 AM Gyan Mishra <[e-mail]> wrote:

> All
>
> I would like to thank the Greg, Tony and Alvaro on their pointers and
> guidance this morning to help move the ball forward with the requirements
> draft.
>

Thank you for your help.


> What I heard from Greg which rang out loud and clear and I mentioned to
> the requirements authors that today we have BIER MPLS where the BIER header
> is encoded into MPLS label stack layer 2 1/2 and we also have non MPLS BIER
> Ethernet ether type 0xAB37.
>

GS - No, the BIER header is NOT 'encoded' in MPLS. MPLS encapsulates the
BIER packet. Yes, the outer label included BIER specific information
(Set-ID) with the intent of streamlining the forwarding process but without
creating a dependency. We mirrored the format for Eth encap to ensure a
single header format for the BIER architecture standard. It's important to
mention there that the process to get to this point involved a 3-day
design-team meeting where we worked hard, transparently and together and
managed to conclude with a solution which resolved any conflict, satisfied
everyone's concerns, and resulted in a clean, consistent message from a
Working Group that was working under the 'experimental' banner. We moved
from 'experimental' to standards track by listening to all the reasons we
were started out as 'experimental' in the first place - ie This is the
forwarding plane and will require new HW so the work from the WG must
justify a new forwarding model and produce clear standards to follow.


> So why do we need non MPLS BIER encapsulation in an IPV6 environment as
> that IPv6 environment can be supported by non MPLS BIER Ethernet.  The case
> and point here is that eventually just as ATM and Frame Relay now live in
> the technology graveyard so will at some point in time as SRv6 matures will
> become the end all be all “core transport” mechanism for all operators.
>

GS - This is an opinion and not a technical argument. BTW, it is MY opinion
that the largest early adopters for BIER will be financial networks, and
more specifically financial trading floors. These networks are dedicated,
high-speed, multipoint networks that will never adopt SRv6. Remember the
IETF is to create standards that work best for everyone, not just to solve
our individual network issues or vision.


> So that being said we need a another encapsulation method to carry BIER ,
> and per RFC 8296 that gap is filled with Non MPLS BIER Ethernet
> encapsulation today which will work for future SRv6 transport once MPLS
> goes by the wayside.
>

GS - Just for clarity, RFC8296 works today regardless of any
perceived transport wars. That's because a proper layered architecture
avoids layer dependencies.


> At the beginning of the presentation Greg corrected me and stated that
> that after the BIERin6 independent model draft was published, that the
> requirement draft came about to build a set of requirements as to the “why”
> we need BIER to work in a non MPLS BIER in an IPv6 environment when we
> already have the BIER Ethernet encapsulation that fits the bill and works.
>

GS - Not exactly... Let me clarify:
  BIERin6 first published in Oct 2017. I think the primary interest at the
time was Homenet. The WG had no issues with the solution, but then neither
had many members stepping up to help. Our AD's (and through extension the
IAB) ingrained in us the need to remain diligent in the publication of our
standards and only progress work that remained consistent with our existing
work and showed community support. The first was true, the second was a
little slight at the time.

  April 2018 - the BIER Encap (misnomer) draft was published. It was
pointed out from the beginning that there is a stark difference between
encap and encode and that the WG has already published standards for
encode. So the question was asked, why do you need a new encoding and
encap. The answer given at the time was that there was a need to address
non-BIER routers in the network and that this solution would be
transitional. The WG pushed back that ANY transition mechanisms MUST NOT
require HW changes. We also heard things like:
"integrated into IPv6" - not true, it's merely BIER information encoded
into extension headers. Operationally this is identical to BIERin6 encap,
except that the former would require new HW, something our AD's made clear
needs to be justified before progressing.
"fast-path" - not true.
I/We continued to ask questions trying to understand what benefits could be
implied by encoding the BIER header into an IPv6 extension header rather
than to just encap the BIER packet? And ultimately what is to be
accomplished with any new encap now that we have RFC8296? The WG decided we
needed a problem statement and requirements draft to guide any work in this
space.

The reqs draft was heavily weighted by BIEREncode draft authors and showed
significant effort to direct the conversation. The WG was largely silent on
any support either way. Mike came to me frustrated that nobody is stepping
up to help. My reply was that if the WG doesn't show interest in the work,
maybe that's your answer. Since then, more authors have been added.


> So that’s the million $$ question we are trying to solve here with the
> requirements draft.
>

GS - Yes, but redefined as per my explanation above.


> As for the IPV6 6MAN questions, I was brought on board by Mike McBride as
> the IPv6 SME as well as multicast SME - but point being member of 6MAN for
> many years so a go between liaison with 6MAN related to any questions
> regarding following the IPV6 specification for extension header usage per
> RFC 8200.  Both solutions drafts had been reviewed by myself and 6MAN and
> no technical issues were found regarding the solutions.
>

GS - Thank you.


> Alvaro mentioned as far as the list of requirements that they were fairly
> basic but maybe needed some more meat behind it such as the “support
> various L2 link types” but we did not specify.  In previous versions we
> stated L2 agnostic and then switched to various but being vague on which
> L2.  Alvaro also mentioned why OAM should be a requirement.  We may want to
> add a sentence on justification as to why we picked BIER IPv6 requirements
> as required versus optional.
>
> We need to add some more meat in the introduction or maybe even a separate
> section as to what gap is being filled by non MPLS BIER in IPv6 environment
> using IPv6 encapsulation and encoding the BIER header versus Non MPLS BIER
> Ethernet.  Also maybe use the requirements section to see if a new
> requirement that maybe a gap that is not covered by non MPLS BIER Ethernet
> that can be covered by non MPLS BIER in an IPv6 environment.
>
> At the end of the call when we rolled through the last two drafts and went
> into overtime I heard the ask for call for adoption for BIERin6 independent
> model.
>

GS - Both v6 draft presentations asked for adoption. One of them had
technical arguments for doing so. The other was list of grievances.


> https://datatracker.ietf.org/doc/draft-zhang-bier-bierin6/
>
> I would not think we are ready to adopt any non MPLS BIER in IPV6
> environment solution if we still do not have the requirements set as to the
> gap that is being filed and problem being solved that cannot be done today
> with non MPLS BIER Ethernet.
>

GS - Agree.


> The flip side of the comment above is that if we are ready to adopt and we
> decided we can skip answering the “why” questions, so then do we need the
> requirements draft at all if that’s the case as we have made the decision
> to go with a single solution and are closing the door on any other
> options.  If the latter then we hang tight on any adoption of any solution
> and wait till the requirements draft is completed and adopted followed by
> moving forward with adopting any solutions.
>

GS - Agree. And my suggestion for new authors on the reqs draft still
stands. I'm not demanding, I'm suggesting, based on the document's history.
Your presentation of the history of the draft made it clear that there has
been little direction, less clarity, and potentially an excessive amount of
motivation to justify one solution over the other, leaving us with the
mushy reqs draft we have today. Let's get this draft right, then we will
have the guidance to address any potential solutions.

Thanks,
Greg


> Kind Regards
>
> Gyan
>
>
> --
>
> <http://www.verizon.com/>
>
> *Gyan Mishra*
>
> *Network Solutions A**rchitect *
>
>
>
> *M 301 502-134713101 Columbia Pike *Silver Spring, MD
>
>