Re: [bess] PFN questions in rfc4732bis

[Greg Mirsky <[e-mail]>]

Hi Jorge,
thank you for your expedient feedback to the proposed updates. I have
several follow-up notes logged below under the GIM>> tag.

Regards,
Greg

On Fri, Mar 22, 2024 at 9:30 AM Jorge Rabadan (Nokia) <
[e-mail]> wrote:

> Hi Greg,
>
>
>
> Thanks for getting back.
>
> My comments in line with [jorge].
>
>
>
> Jorge
>
>
>
> *From: *Greg Mirsky <[e-mail]>
> *Date: *Thursday, March 21, 2024 at 2:41 AM
> *To: *[e-mail] <
> [e-mail]>, [e-mail] <
> [e-mail]>, MPLS Working Group <
> [e-mail]>, [e-mail] <[e-mail]>
> *Subject: *PFN questions in rfc4732bis
>
>
>
> *CAUTION:* This is an external email. Please be very careful when
> clicking links or opening attachments. See the URL nok.it/ext for
> additional information.
>
>
>
> Dear All,
>
> following the presentation of our work on the Post-stack First Nibble
> (PFN) to the BESS WG at IETF-119, I took an AP to come with questions and
> proposals for the authors of rfc4732bis. I thought that once the authors of
> the respective drafts converge on the updates, we share them with the BESS
> WG. Below, please find my notes:
>
> ·         rfc4732bis recognizes MPLS Entropy label as a source of entropy
> for load-balancing. 1stnibble draft also refers to RFC 6391
> <https://datatracker.ietf.org/doc/rfc6391/> as another optional source of
> the entropy for load-balancing. Would it be helpful adding a refgerence to
> RFC 6391 in the discussion of the load-balancing in rfc4732bis?
>
> [jorge] that should be fine.
>
> ·         The definition of the C flag in Section 7.11 is as follows:
>
>        C        If set to 1, a control word [RFC4448] MUST be present
>
>                 when sending EVPN packets to this PE.  It is
>
>                 recommended that the control word be included in the
>
>                 absence of an entropy label [RFC6790].
>
> To reflect the position expressed in the 1stnibble draft, perhaps the
> following update is appropriate:
>
> NEW TEXT:
>
>        C        If set to 1, a control word [RFC4448] MUST be present
>
>                 when sending EVPN packets to this PE.
>
> END
>
> [jorge] I personally see no issue with the above update if the other
> co-authors are ok too.
>
> ·         Furthermore, the following update may be considered in Section
> 7.11.1:
>
> OLD TEXT:
>
>    *  per-ESI-and-EVI attributes P, B are conveyed, and;
>
>
>
>    *  per-EVI attributes MTU, Control Word and Flow Label MUST be zero.
>
> NEW TEXT:
>
>    *  per-ESI-and-EVI attributes P, B are conveyed;
>
>
>
>    *  per-EVI attributes MTU, and Flow Label MUST be zero, and;
>
>
>
>    *  per-EVI attribute Control Word MUST be set.
>
> [jorge] I don’t think the above change is correct. The reason is this
> refers to the L2 attributes extended community sent along with the AD per
> EVI routes in EVPN ELAN services (not EVPN VPWS), and this route is purely
> used for multi-homing. The non-multihoming related attributes MUST be zero,
> since they are signaled along with the inclusive multicast ethernet tag
> route. So the current text is correct.
>
GIM>> I acknowledge that the use of the Control Word in this scenario that
I am not familiar with. The motivation for proposing this update is based
on my itnerpretation of the definition of the C flag in Section 7.11 (see
above) and the intention of the draft-ietf-mpls-1stnibble to deprecate the
use of non-PSH encapsulations of non-IP payload. As I understand the
current text in question, by setting the Control Word to zero the control
plane will produce packets without PSH (PW CW). If that is correct, then
that is precisely what our work on the PFN aims to exclude.

>
>
> ·         The text in Section 18, following the first paragraph, may be
> replaced with the following text:
>
> NEW TEXT:
>
> In order to avoid frame misordering described in the above paragraph,
>
> and following conclusions of [I-D.ietf-mpls-1stnibble], the control word
>
> MUST be used in all use cases.
>
> END
>
> [jorge] so basically you are suggesting to use CW always and use MUST as
> normative language, irrespective of a) the underlaying transport, b)
> whether entropy label is used and c) whether deep packet inspection for
> ECMP is used. The only problem that I see is that, till now,
> implementations not supporting CW were still compliant with RFC7432 and
> this bis draft. Now it would not be the case anymore.
>
> I personally think it might be better to use a SHOULD, e.g.:
>
> NEW:
>
> In order to avoid frame misordering described in the above paragraph,
>
> the control word SHOULD be used in all use cases
>  [I-D.ietf-mpls-1stnibble].
>
> END
>
>
>
> But I’d like to hear from other coauthors and the BESS WG to see if this
> is ok.
>
> Finally, if we were to add [I-D.ietf-mpls-1stnibble] as a reference, it
> would be an informative reference.
>
GIM>> We've used SHOULD in RFC 8469
<https://datatracker.ietf.org/doc/html/rfc8469>, and believe that it is
time to deprecate the practice of non-PSH encapsulation of non-IP payload
in MPLS altogether, and use PSH/CW in all scenarios for non-IP payload.
I'll note that the deprecation, as defined in draft-ietf-mpls-1stnibble,
applies only to new deployments and implementations. (I believe that an
intellegent implementation that supports non-PSH(non-CW) encapsulation is
also capable of using PW CW encapsulation.)

>
>
> Please share you questions, comments, and suggestions.
>
>
>
> Regards,
>
> Greg
>