Re: [lisp] Mirja Kühlewind's Discuss on draft-ietf-lisp-rfc6830bis-16: (with DISCUSS and COMMENT)

[Spencer Dawkins at IETF <[e-mail]>]

I haven't balloted on this document yet, but since I will, and would love
to ballot No-Objection ...

On Thu, Sep 20, 2018 at 5:58 AM Mirja Kuehlewind (IETF) <[e-mail]>
wrote:

> Hi Dino,
>
> it’s fine with me to leave the details to rfc6040 but then also the
> details in the current text should be removed (because giving only half the
> details really doesn’t seem right).
>
> However, I totally disagree with your comment on providing details that
> are not implemented. If they are not implemented correctly, it might even
> be more important to spell them out in this document, so implementors have
> chance to update their (future) implementation to do the correct thing.
> Having deployed implementations that are non standard-conform always
> happens and in this case it is probably not specifically problematic as it
> doesn’t impact interoperability. However, it is important though that the
> spec is correct.
>

Tl;dr - you're both right. Do the right thing.

If I'm following this conversation correctly, the situation is

- IP fragmentation can be problematic, roughly as a function of the number
of fragments that any given IP packet is fragmented into, but
- most deployed LISP implementations don't do path MTU discovery now

Is that it?

If so, what I'd suggest, is actually saying it that way.

Mirja's right, that we're advising people to do path MTU discovery of some
type (and I'm pretty sure https://tools.ietf.org/html/rfc4821#section-10.3
is what we would recommend in this case).

Dino's right, that putting out a standard that doesn't reflect deployed
implementations won't inspire people to conform to standards.

But do the right thing, of course ...

Spencer


> Mirja
>
>
> > Am 18.09.2018 um 18:56 schrieb Dino Farinacci <[e-mail]>:
> >
> > As I already said, this text is too detailed and repeats what is in
> other RFCs. And since implementations do what they already do, adding more
> details that are not implemented, IMO, is not good form.
> >
> > Dino
> >
> >> On Sep 18, 2018, at 3:32 AM, Mirja Kuehlewind (IETF) <
> [e-mail]> wrote:
> >>
> >> Hi Dino,
> >>
> >> please see below.
> >>
> >>> Am 17.09.2018 um 19:48 schrieb Dino Farinacci <[e-mail]>:
> >>>
> >>>> PROPOSED
> >>>> "The 'Explicit Congestion Notification' (ECN) field (bits 6 and 7
> >>>>   of the IPv6 'Traffic Class' field) [RFC3168] requires special
> treatment in
> >>>>   order to preserve the use of ECN on the path.
> >>>>   ITR encapsulation MUST copy the 2-bit 'ECN' field from the inner
> >>>>   header to the outer header, inline with the ’Normal Mode’ in
> section 4.1
> >>>>   of [RFC6040].  Re-encapsulation SHOULD follow the decapsulation as
> described
> >>>>   below and then 2-bit 'ECN' field from the stripped inner header to
> the
> >>>>   new outer header.“
> >>>
> >>> I did not include this text because the last sentence is not formed
> well. Please restate. A verb is missing.
> >>
> >> copy
> >>
> >>>
> >>>> "The 'Explicit Congestion Notification' (ECN) field (bits 6 and 7
> >>>>   of the IPv6 'Traffic Class' field) requires special treatment on
> >>>>   decapsulation in
> >>>>   order to avoid discarding indications of congestion,
> >>>>   inline with section 4.2 of [RFC6040]. If
> >>>>   the 'ECN‘ field of the outer header contains a congestion
> indication
> >>>>   codepoint (the
> >>>>   value is '11', the Congestion Experienced (CE) codepoint) and the
> inner
> >>>>   header indicates ECN support (either ECT(0) or ECT(1) codepoint is
> set),
> >>>>   then ETR decapsulation MUST also set CE field in the inner header
> that is
> >>>>   used
> >>>>   to forward the packet beyond the ETR. If the inner packet is marked
> as non-
> >>>>   ECT but the outer header has the CE mark set, the packet MUST be
> dropped
> >>>>   instead. Any discrepancy between the inner and outer header for
> non-ECT,
> >>>>   ECT(0) and ECT(1) MUST NOT be copied from the outer header. These
> >>>>   requirements preserve
> >>>>   CE indications when a packet that is ECN-capable traverses a LISP
> tunnel
> >>>>   and becomes marked with a CE indication due to congestion between
> >>>>   the tunnel endpoints or transforms an CE into loss if that packet
> is not
> >>>>   ECN-capable conserving the congestion indication towards a non-ECN
> enables
> >>>>   endpoint.”
> >>>
> >>> I didn’t include this text because (1) it under states what to do with
> IPv4, (2) it has too much detail that is already in RFC6040, and (3) it
> undoes text that other reviewers have offered.
> >>
> >> I didn’t change the mentioning of IPv6 here. Yes please at IPv4.
> >>
> >> You can remove all this text and only point to rfc6040. That would
> actually my preferred solution. I don’t think it „undoes“ text; it just
> adds what was missing in compliance with RFC6040. Anyway it doesn’t matter
> point being that it should specify the same things as RFC6040 does not
> matter what other have ofter because RFC6040 is the IETF-consensus doc how
> describing how to handle this.
> >>
> >>>
> >>>
> >>>> Please also remove the duplicated text after these bullet lists in
> the draft!
> >>>
> >>> You have to tell me what text. I am too confused at this point on what
> you want.
> >>
> >> This is the text in the en-/ and decapsulation lists:
> >>
> >> "The 'Explicit Congestion Notification' (ECN) field (bits 6 and 7
> >>     of the IPv6 'Traffic Class' field) requires special treatment in
> >>     order to avoid discarding indications of congestion [RFC3168].
> >>     ITR encapsulation MUST copy the 2-bit 'ECN' field from the inner
> >>     header to the outer header.  Re-encapsulation MUST copy the 2-bit
> >>     'ECN' field from the stripped outer header to the new outer
> >>     header."
> >>
> >> "The 'Explicit Congestion Notification' (ECN) field (bits 6 and 7
> >>     of the IPv6 'Traffic Class' field) requires special treatment in
> >>     order to avoid discarding indications of congestion [RFC6040].  If
> >>     the 'ECN' field contains a congestion indication codepoint (the
> >>     value is '11', the Congestion Experienced (CE) codepoint), then
> >>     ETR decapsulation MUST copy the 2-bit 'ECN' field from the
> >>     stripped outer header to the surviving inner header that is used
> >>     to forward the packet beyond the ETR.  These requirements preserve
> >>     CE indications when a packet that uses ECN traverses a LISP tunnel
> >>     and becomes marked with a CE indication due to congestion between
> >>     the tunnel endpoints."
> >>
> >> And this text comes up right after the list in the same section:
> >>
> >> "The Explicit Congestion Notification ('ECN') field occupies bits 6
> >>  and 7 of both the IPv4 'Type of Service' field and the IPv6 'Traffic
> >>  Class' field [RFC6040].  The 'ECN' field requires special treatment
> >>  in order to avoid discarding indications of congestion [RFC6040].  An
> >>  ITR/PITR encapsulation MUST copy the 2-bit 'ECN' field from the inner
> >>  header to the outer header.  Re-encapsulation MUST copy the 2-bit
> >>  'ECN' field from the stripped outer header to the new outer header.
> >>  If the 'ECN' field contains a congestion indication codepoint (the
> >>  value is '11', the Congestion Experienced (CE) codepoint), then ETR/
> >>  PETR decapsulation MUST copy the 2-bit 'ECN' field from the stripped
> >>  outer header to the surviving inner header that is used to forward
> >>  the packet beyond the ETR.  These requirements preserve CE
> >>  indications when a packet that uses ECN traverses a LISP tunnel and
> >>  becomes marked with a CE indication due to congestion between the
> >>  tunnel endpoints."
> >>
> >> The last text bit does not add any information; it just states all
> normative requirement twice, even using basically exactly the some words.
> This can lead to discrepancies and it really not necessary. I’d recommend
> to just remove the last text block here (and fix the IPv6/IPv4 issue in the
> other blocks).
> >>
> >> Mirja
> >>
> >>
> >>>
> >>>> Further I believe my discuss points 2) and 4) are not fully resolved
> yet. Also I would like to at least see more explanation about the approach
> for extensibility that was taken in this doc (point 6).
> >>>
> >>> You are going to have to repeat what they are because too many emails
> have flown by since your initial post. And for extensibility, we discuss it
> in RFC8060 and don’t think anything more should be said here otherwise, we
> will duplicate unnecessary text.
> >>>
> >>> Another new diff file enclosed.
> >>>
> >>> Dino
> >>>
> >>>
> >>>
> >>>
> >>>
> >>>
> >>
> >
>
>