Configuration Guidelines for DiffServ Service Classes
RFC 4594

[Ballot discuss]
I object to the normative language in this document while it is just informational instead of BCP and didn't get real operator review as far
as I understand.

---
See below for the review I received by Pekka Savola from the Ops Directorate:

Well, I guess the doc would be much better read if it was 20 pages
instead of 60.

Certainly, if you look at the document writeup, I don't see any
operators acknowledged there...

I did take a very quick look at it; my main concern was that the doc
used uppercase keywords in what seemed to be an inappropriate fashion,
_especially_ for an Informational document, for example:

3.1.  Current Practice in The Internet
...
  RECOMMENDED Network Edge Conditioning for CS7 DSCP marked packets:
  o  Drop or remark CS7 marked packets at ingress to DiffServ network
      domain.
  o  CS7 marked packets SHOULD NOT be sent across peering points.
      Exchange of control information across peering points SHOULD be
      done using CS6 DSCP, using Network Control service class.
While this kind of language (all over the document) might or might not
be appropriate in a BCP document which has firmly been established in
the operator field, I'm somewhat concerned to see it here..

It's not clear what should be the constructive next steps though. More
discussion and participation from the operational fora wouldn't hurt,
but it isn't clear whether requiring such would be reasonable.

In any case, I think the use of uppercase keywording requires some
serious thinking and/or rewording.

[Ballot discuss]
in Section "3.3 OAM Service Class" I see:

  Traffic characteristics:
  o  Variable size packets (50 to 1500 bytes in size)
  o  Intermittent traffic flows
  o  Traffic may burst at times
  o  Both elastic and inelastic flows
  o  Traffic not sensitive to delays

As protocols you list for exampel SNMP, COPS.
Well, SNMP can have packets larger than 1500.
Certainly when you do SNMP over TCP (RFC3439).
And COPS is over TCP, and typically would use larger packets I think.

NETCONF soon will be deployed too.
And again it may very well use pakets larger than 1500.

[Ballot discuss]
The draft says:

  For VoIP (telephony) service, a common approach is to use signaling
  protocols such as SIP, H.323, H.248, MEGACO, RSVP, etc. to negotiate
  admittance and use of network transport capabilities.  When a user
  has been authorized to send voice traffic, this admission procedure
  has verified that data rates will be within the capacity of the
  network that it will use.  Since RTP voice does not react to loss or
  delay in any substantive way, the network SHOULD police at ingress to
  ensure that the voice traffic stays within its negotiated bounds.
  Having thus assured a predictable input rate, the network may use a
  priority queue to ensure nominal delay and variation in delay.

There are certainly cases where inelastic codecs mean "RTP voice does not react to loss
or delay in any substantive way", but that is also certainly not always the
case.  Using that as a justification for this SHOULD seems problematic.  There
are also cases where admission control decisions are more fine grained than this implies;
the authors may wish to consider whether to consider the question of admission
control more thoroughly or simply drop back to a position which describes what
happens when admission control is in place.

[Ballot comment]
Kathie Nichols and Charlie Liu are mentioned in the summary of changes
but should probably be added to the Acknowledgements.

Noted by Spencer Dawkins (Gen-ART reviewer) for copy editors: There are still a few items like "due to the vest number" (in Section 2.2, should be "due to the vast number").

This version has addressed WGLC call comments including comments which
came in when we sent request to SIP, SIPPING, MPLS, AVT, DCCP and other
WGs...  The 00->01 addressed three invited, detailed "Early Reviews" by
David Black, Brian Carpenter, and Alan O'Neill.

Review telechat to handle Last Call comments yesterday - finishing touches
on a draft of -02.  All authors present along with David Black as
the major Last Call commenter and TSV Doctor

Waiting..... ..... ..... for Kwok to do a rev for David Black's proposal
to address the big multi-WGLC comment - one major comment set came in
that needs addressing.  We are eager to advance.  Email has gone to Kwok
about three times now.

[Note]: 'WGLC starts - forwarding the WGLC to multiple other groups too who need to review applicability -
therefore it''s a four week LC. ' added by Allison Mankin

[Note]: 'WGLC starts - forwarding the WGLC to multiple other groups too who need to review applicability -
therefore it''s a four week LC. ' added by Allison Mankin