Re: [MBONED] Benjamin Kaduk's No Objection on draft-ietf-mboned-deprecate-interdomain-asm-06: (with COMMENT)

[Tim Chown <[e-mail]>]

Hi,

On 9 Jan 2020, at 02:54, Benjamin Kaduk via Datatracker <[e-mail]<mailto:[e-mail]>> wrote:

Benjamin Kaduk has entered the following ballot position for
draft-ietf-mboned-deprecate-interdomain-asm-06: No Objection

When responding, please keep the subject line intact and reply to all
email addresses included in the To and CC lines. (Feel free to cut this
introductory paragraph, however.)


Please refer to https://www.ietf.org/iesg/statement/discuss-criteria.html
for more information about IESG DISCUSS and COMMENT positions.


The document, along with other ballot positions, can be found here:
https://datatracker.ietf.org/doc/draft-ietf-mboned-deprecate-interdomain-asm/



----------------------------------------------------------------------
COMMENT:
----------------------------------------------------------------------

Thanks for this well-written document!

Section 2.1

  to deliver traffic from the sender(s) to the receivers.  If there are
  multiple senders for a given group, traffic from all senders will be
  delivered to the receiver.  Since receivers specify only the group

nit: "receivers" plural.

Fixed.

Section 3.1

  troubleshoot.  Some of these issues include complex flooding Reverse
  Path Forwarding (RPF) rules, state attack protection, and filtering
  of undesired sources.

I'm not sure how to parse "complex flooding Reverse Path Forwarding
(RPF) rules" (but think I could figure out "complex Reverse Path
Forwarding (RPF) flooding rules" -- do I need to try harder?

I’ve simply omitted flooding as it’s implied, I think.

  within one domain.  While this approach solves the MSDP issues, it
  does not solve the problem of unauthorised sources sending traffic to
  ASM multicast groups; this security issue is one of biggest problems
  of interdomain multicast.

Perhaps it's worth expanding on the security issue as including a
substantial level of traffic amplification.  (Or do I misunderstand the
security issue?)

The issue here is injection of spoofed traffic.  I’m not sure it’s amplification in the sense used for say open DNS resolver amplification, as the receivers and links to receivers only see one copy of whatever is injected.

Section 4.1

  The more inclusive interpretation of this recommendation is that it
  also extends to deprecating use of ASM in the case where PIM is

Are we claiming or disclaiming this "more inclusive interpretation"?
The current wording feels ambiguous, and I don't think we should leave
it that way.

We discussed this previously in a WG meeting iirc and it was left that way as there was a feeling that within a single domain an operator can in practice do as they please, and the real focus is in deprecating inter-domain ASM.

Section 4.2

  support SSM (i.e., explicitly sending source-specific reports).  The
  updated IPv6 Node Requirements RFC [I-D.ietf-6man-rfc6434-bis] states

nit: the I-D reference is not yet an RFC, so it's not proper to refer to
it as one; since it's only an informative reference we will not
necessarily wait to publish this document until that one has an RFC
number assigned, so I think this should be reworded.  (Also, even if
there was a new RFC number, "updated [...] RFC [RFCXXXX]" would still
scan kind of strangely.)

This is now RFC 8504, updated in the text.

Section 4.3

  The recommendation to use SSM for interdomain multicast means that
  applications should properly trigger the sending of IGMPv3/MLDv2
  source-specific report messages.  It should be noted, however, there
  is a wide range of applications today that only support ASM.  In many
  cases this is due to application developers being unaware of the
  operational concerns of networks.  This document serves to provide
  clear direction for application developers to support SSM.

Without some discussion of the relative level of difficulty for applications
tosupport SSM (to compare against the well-portrayed problems with network
support for ASM), it would be fairly easy to misread this as network
operators getting an IETF BCP to try to force applications to do what's easy
for the network.  I'm given to understand that's not actually the case here,
as the changes in application software are fairly minor, and no one would
expect existing applications to grow a way to announce multicast source
addresses spontaneously, so a little more text might go a long way.

I’ve added/changed to:

"This document serves to
        provide clear direction for application developers who might currently only
        consider using ASM to instead support SSM, which only requires relatively minor
        changes for many applications, particularly those with single sources.”


Section 4.4

  While the WG discussions had consensus that best practices should be
  developed to explain when to use SSM in applications, e.g, when ASM
  without (S,G) state in the network is better, or when dedicated
  service-discovery mechanisms should be used, it was agreed that
  documenting such practices are outside the scope of this document.

It might be possible to rephrase this in terms of "the conclusions of
the MBONED WG" or "the consensus of the MBONED WG"; the current phrasing
feels a little informal to me, though I don't place much weight on my
sentiment here.

Now:

"While the consensus of the MBONED WG discussions was that best practices should be developed
        to explain when to use SSM in applications,
        e.g, when ASM without (S,G) state in the network is better, or when dedicated
        service-discovery mechanisms should be used, the WG also agreed that documenting
        such practices are outside the scope of this document.”


Section 4.6

A lot of this section feels somewhat speculative and leaves me uncertain
what the BCP recommendations in this space are.

  In the case of existing ASM applications that cannot readily be
  ported to SSM, it may be possible to use some form of protocol
  mapping, i.e., to have a mechanism to translate a (*,G) join or leave
  to a (S,G) join or leave, for a specific source, S.  The general

nit: I think this would read better with fewer commas, i.e., "to
translate a (*,G) join or leave to a (S,G) join or leave for a specific
source S".

Fixed.

Section 4.9

  it.  This allows easy migration of ASM applications to SSM/PIM-SSM
  solely through application side development to handle source-

nit: hyphenate "application-side".

Fixed.

  signaling via IGMPv3/MLDv2 and using SSM addresses.  No network

Do the application developers also consider this work "easy"? ;)

The word “easy” has been delated, but “simply” remains :)

"This allows migration of ASM applications
        to SSM/PIM-SSM solely through application-side development
        to handle source-signaling via
        IGMPv3/MLDv2 and using SSM addresses."

  When running PIM-SM, IGMPv3/MLDv2 (S,G) membership reports may also
  result in the desired PIM-SSM (S,G) operations and bypass any RP
  procedures, but this is not standardized but depends on
  implementation and may require additional configuration in available
  products.  [...]

nit: this sentence is fairly long and convoluted; could it be reworded
and/or split up into smaller pieces?

Good suggestion.

"When running PIM-SM, IGMPv3/MLDv2 (S,G) membership reports may also result in
        the desired PIM-SSM (S,G) operations and bypass any RP procedures. This is not
        standardized but depends on implementation and may require additional configuration
        in available products."

  Note that these migration recommendations do not include the
  considerations when or how to evolve those intradomain applications
  best served by ASM/Bidir-PIM from PIM-SM to Bidir-PIM.  This may also
  be important but is outside the scope of this document.

nit: is there a missing word ("for"/"about"/"on") after
"considerations"?

"Note that these migration recommendations do not include considerations on when
        or how to evolve”


  Accordingly, this document recommends that future work for general
  purpose interdomain IP multicast focus on SSM items listed in
  Section 4.

nit: hyphenate "general-purpose".

Done.

Section 6

We could perhaps consider mentioning again the consequences of
infrastructure assuming that SSM-range addresses are always used for
SSM, such as during the transition period where applications migrating
from ASM to SSM continue to use ASM-range addresses, as mentioned in
Section 4.7.

Changing the model for intradomain multicast to one where
source discovery/state-propagation is not done by the network and is
instead a responsibility of the application layer means that the
applications are responsible for properly implementing such
functionality.  While it's true that we do want applications to not have
bugs in general (and that would hopefully go without saying), we might
want to discuss the scope of potential issues if applications get this
wrong.  Could application bugs (or bugs in host-level IGMPv3 or MLDv2
implementations) specific to SSM usage lead to any worse problems than
just "application traffic doesn't flow", e.g., state exhaustion in the
network?

I’m not sure what text you’d like to see added, so as yet have not added any.

This could be the subject of an additional draft providing more specific guidance for developers.

Section 10.1

I'm not seeing any references to RFC 4610 that would qualify it as
normative; if normative is indeed the proper status, perhaps additional
citations are in order?

Moved :)

Many thanks,
Tim