Re: [MEDIACTRL] Review of draft-ietf-mediactrl-call-flows-10

worley@ariadne.com (Dale R. Worley) Mon, 15 April 2013 21:50 UTC

Return-Path: <worley@shell01.TheWorld.com>
X-Original-To: mediactrl@ietfa.amsl.com
Delivered-To: mediactrl@ietfa.amsl.com
Received: from localhost (localhost [127.0.0.1]) by ietfa.amsl.com (Postfix) with ESMTP id 66FCB21F9408 for <mediactrl@ietfa.amsl.com>; Mon, 15 Apr 2013 14:50:05 -0700 (PDT)
X-Virus-Scanned: amavisd-new at amsl.com
X-Spam-Flag: NO
X-Spam-Score: -1.286
X-Spam-Level:
X-Spam-Status: No, score=-1.286 tagged_above=-999 required=5 tests=[AWL=-0.206, BAYES_00=-2.599, FR_DOT_FEVER_5=3.9, GB_I_LETTER=-2, RCVD_IN_DNSWL_LOW=-1, RCVD_IN_SORBS_WEB=0.619]
Received: from mail.ietf.org ([12.22.58.30]) by localhost (ietfa.amsl.com [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id bKH2mISByR5w for <mediactrl@ietfa.amsl.com>; Mon, 15 Apr 2013 14:50:03 -0700 (PDT)
Received: from TheWorld.com (pcls6.std.com [192.74.137.146]) by ietfa.amsl.com (Postfix) with ESMTP id CADB321F93F9 for <mediactrl@ietf.org>; Mon, 15 Apr 2013 14:50:02 -0700 (PDT)
Received: from shell.TheWorld.com (root@shell01.theworld.com [192.74.137.71]) by TheWorld.com (8.14.5/8.14.5) with ESMTP id r3FLmLw6024615 for <mediactrl@ietf.org>; Mon, 15 Apr 2013 17:48:23 -0400
Received: from shell01.TheWorld.com (localhost.theworld.com [127.0.0.1]) by shell.TheWorld.com (8.13.6/8.12.8) with ESMTP id r3FLmKaM2683903 for <mediactrl@ietf.org>; Mon, 15 Apr 2013 17:48:21 -0400 (EDT)
Received: (from worley@localhost) by shell01.TheWorld.com (8.13.6/8.13.6/Submit) id r3FLmKaO2695748; Mon, 15 Apr 2013 17:48:20 -0400 (EDT)
Date: Mon, 15 Apr 2013 17:48:20 -0400 (EDT)
Message-Id: <201304152148.r3FLmKaO2695748@shell01.TheWorld.com>
From: worley@ariadne.com (Dale R. Worley)
Sender: worley@ariadne.com (Dale R. Worley)
To: mediactrl@ietf.org
In-reply-to: <97700AB5-E682-4DA2-BCD6-874A50F66575@standardstrack.com> (eburger@standardstrack.com)
References: <201212192009.qBJK9j8a3356324@shell01.TheWorld.com> <20121226131152.4660538e@rainpc> <97700AB5-E682-4DA2-BCD6-874A50F66575@standardstrack.com>
Subject: Re: [MEDIACTRL] Review of draft-ietf-mediactrl-call-flows-10
X-BeenThere: mediactrl@ietf.org
X-Mailman-Version: 2.1.12
Precedence: list
List-Id: Media Control WG Discussion List <mediactrl.ietf.org>
List-Unsubscribe: <https://www.ietf.org/mailman/options/mediactrl>, <mailto:mediactrl-request@ietf.org?subject=unsubscribe>
List-Archive: <http://www.ietf.org/mail-archive/web/mediactrl>
List-Post: <mailto:mediactrl@ietf.org>
List-Help: <mailto:mediactrl-request@ietf.org?subject=help>
List-Subscribe: <https://www.ietf.org/mailman/listinfo/mediactrl>, <mailto:mediactrl-request@ietf.org?subject=subscribe>
X-List-Received-Date: Mon, 15 Apr 2013 21:50:05 -0000

> From: Eric Burger <eburger@standardstrack.com>;
> 
> Are we done yet?

Looking over the messages, I see that I should have responded to
Lorenzo's message of 26 Dec
(http://www.ietf.org/mail-archive/web/mediactrl/current/msg01781.html).
Of course, most of the items have already been resolved, but there are
a few that need more discussion.

> > item 1) various
> > 
> > The indentation of the header lines identifying various messages is
> > inconsistent.  I can find the following lines in various places:
> [...]
>
> [LM] You're right, the indentation often varies, and the logic behind
> this currently is according to whether or not the example fits the page.
> Instructions about this would be welcome!

I think leaving it to the Editor may be the best, and the Editor will
have to worry about fitting the lines on the page anyway.  The Editor
probably as style conventions about indenting.

> > item 2) various
> > 
> > There are a lot of places where an arrow seems to go from a network
> > entity to itself, such as "create conf and its ID" below:
> > 
> >                     AS                                 MS
> >                     |                                  |
> >                     | 1. CONTROL (create conference)   |
> >                     |++++++++++++++++++++++++++++++++>>|
> >                     |                                  |--+ create
> >                     |                                  |  | conf and
> >                     |       2. 200 OK (conferenceid=Y) |<-+ its ID
> >                     |<<++++++++++++++++++++++++++++++++|
> 
> [LM] Yes, in that case the arrow was not meant as a network connection,
> but as a further step in the sequence diagram, that is what that
> CONTROL message triggers in the MS application logic. The same applies
> to all the examples where this happens. If you think the arrowless
> diagram is clearer to the reader we might remove them: the problem with
> this is that, without arrows, some diagrams might be harder to read,
> i.e., in terms of who does what in diagrams with several components
> interacting with each other. As an alternative, we might clarify what
> those arrows are and what they are not. What do you think about this?

We might be able to choose a syntax that less resembles the
propagation of a message.  Perhaps this:

                     AS                                 MS
                     |                                  |
                     | 1. CONTROL (create conference)   |
                     |++++++++++++++++++++++++++++++++>>|
                     |                                  |-- create
                     |                                  |   conf and
                     |                                  |   its ID
                     |       2. 200 OK (conferenceid=Y) |
                     |<<++++++++++++++++++++++++++++++++|

That makes it clear that the MS is performing an action, but it does
not look like generating a message.

> > item 7) section 4
> > 
> > This section starts with a paragraph over one page long.  Could it be
> > broken into multiple paragraphs?
> 
> [LM] Did you mean section 4.1? Section 4 starts with just a few lines.

Yes, you are right, I meant section 4.1.

> > item 9) section 4 figure 1
> > [...]
> > - The states "'202' sent" and "'update' sent" have the same outgoing
> >   transitions, so they should probably be merged.
> 
> [LM] I'm not sure about this: 202 is only sent after the first CONTROL
> message to tell the AS it may take a while to process the request, after
> which you can either have further 202 messages should the processing
> take even more time, or one or more UPDATE messages related to that
> CONTROL request after that. They're two different messages in the
> protocol semantics. Nevertheless, it's also true that, according to
> RFC6230, a 202 is basically a pre-REPORT message... any feedback on
> this?

The idea is that the MS is in the same state after sending the 202 as
it is after sending a subsequent update, because what it can do next
is the same, and so at least theoretically, the two states do not need
to be distinguished.

My idea is that the diagram could be simplified to look like this:

   +------------------+  CONTROL/-  +------------------+ API 202/202
   | Idle/'terminate' |------------>| CONTROL received |---------+
   +------------------+             +------------------+         |
     ^          ^   ^                  |     |                   |
     |          |   |   API 200/200    |     |                   |
     |          |   +------------------+     |                   |
     | 200/-    |      API Error/Error       |                   |
     |          +----------------------------+                   |
     |                                                           |
     |                                                           |
     |                      API terminate/                       v
   +------------------+     REPORT terminate      +-------------------+
   | 'terminate' sent |<--------------------------| command executing |
   +------------------+                           +-------------------+
                                                       ^        |     
                                                       |        | API update/
                                                       |        | REPORT update
                                                       |        |
                                                       |        |
                                                       |        |
                                                 200/- |        | 
                                                       |        v
                                                     +---------------+
                                                     | 'update' sent |
                                                     +---------------+

> > item 9) section 4 figure 1
> > [...]
> > - What happens if the API detects an error after 202 has been sent?
> >   There is no "API Error" arc from "'202' sent"/"'update' confirmed".
> 
> [LM] Good point. The idea is that, once the MS sends a 202, the request
> has been correctly processed, and so is supposed to be fine as far as
> the protocol is concerned. I guess there could be circumstances when an
> errors occurs after that, e.g., in retrieving an audio file to be played
> to a UAC: in that case, the AS would be informed of the termination of
> the dialog/request in a REPORT terminate, including details on the issue
> which are Control Package specific. Do you think explaining this in the
> text may help readers?

I think an explanation in the text would be helpful, although if the
reader has a thorough understanding of what information is carried in
REPORT terminate they would deduce the answer.  Perhaps this
explanation would help those who are not completely familiar:

   The MS may send a 202 response after it determines that the request
   request contains no errors that cannot be reported in a later REPORT
   terminate request.  After the MS sends a 202 response, any error that
   it (or the API) finds in the request is reported in the final REPORT
   terminate request.

> > item 12) section 5
> > 
> > The term "COMEDIA" is used, and a reference is given to RFC 4145.  But
> > that RFC does not use the term "COMEDIA" at all.  The framework I-D
> > also uses "COMEDIA" without explicitly defining it.  It would be
> > helpful to insert here a definition of "COMEDIA", which seems to be
> > synonymous with "media of type application/cfw".
> 
> [LM] I thought COMEDIA actually came from RFC4145, standing for
> "connection-oriented media". Even the MSRP RFC refers to it that way. If
> that's incorrect or incomplete we'll fix this in the next version.

I've searched through all the RFCs for the term "comedia".  (The
results are summarized at the end of this message.)  What seems to
have happened is that "comedia" has become an abbreviation for
"connection-oriented media" (i.e., TCP and TLS) and also for two RFCs
regarding support for connection-oriented media (RFC 4145 and RFC
4572).

The problem is that the term seems to be very common among a small
audience centered on Mediactrl, but only among that audience, and
people inside that audience do not take care to define it clearly for
the benefit of people outside that audience.  For instance, RFC 5567
is "An Architectural Framework for Media Server Control" and uses
"comedia" in "... exploit floor control by means of a
Connection-Oriented Media (COMEDIA) [RFC4145] negotiation." but
"comedia" is not listed in its section 2, "Terminology".

I think this would be solved if "comedia" was listed in section 3,
"Terminology":

   COMEDIA:  connection-oriented media (i.e., TCP and TLS).  Also used to
      signify the support in SDP for connection-oriented media, and the
      RFCs that define that support (RFC 4145 and RFC 4572).

> > item 15) section 5.1
> > > [...]
> > This confuses me:  Either the list of packages in the SDP offer/answer
> > is significant or it is not.  If it is not, why is it provided?
> > [...]
> 
> [LM] That section refers to a draft Chris submitted to the MMUSIC WG
> some time ago:
> 
> http://tools.ietf.org/html/draft-boulton-mmusic-sdp-control-package-attribute-07
> 
> so any clarification about the purpose and constraints of such a
> mechanism should be addressed in the specification. Chris, do you think
> this behaviour could be clarified in the draft, in order to have it
> reflected in the Call Flows as a consequence?

This looks complicated.  It seems that
draft-boulton-mmusic-sdp-control-package-attribute has not been
adopted as a working group draft and has expired.  If that is the
case, there will be no definition of the ctrl-package attribute, and
it shouldn't be used in the call-flows.

But if that is fixed, it seems that we can copy some text from
draft-boulton-mmusic-sdp-control-package-attribute to clarify this
paragraph.  I suggest splitting the paragraph, and amending the second
half like this:

   Note that the provided example also includes the indication, from
   both the AS and the MS, of the supported control packages.  This is
   achieved by means of a series of 'ctrl-package' attributes as
   specified in [I-D.boulton-mmusic-sdp-control-package-attribute].  In
   the example, the AS supports (or is only interested to) two packages:
   IVR (Interactive Voice Response) and Mixer (Conferencing and
   Bridging).  The MS replies with the list of packages it supports,
   adding a dummy example package to the list provided by the AS.

   It is worth noting that this exchange of information is only
   advisory and neither endpoint is committing to support the packages
   that are listed or to not supporting the packages that are not
   listed.  The actual negotiation of control packages is done
   subsequently through the use of the framework SYNC transaction
   within the control channel.  Nonetheless, the use of 'ctrl-package'
   attributes can be beneficial both from a resource allocation
   perspective and it could also result in two clients identifying
   they are not capable of successful control channel interactions.
   This would lead to early abandonment of control channel setup (for
   example, a client may choose to terminate the associated SIP dialog
   and not attempt to make the connection).

> > item 17) section 5.3
> > 
> >    In the previous section it has been described how a timeout value for
> >    the keep alive mechanism is negotiated.  Specifically, in the example
> >    the AS and the MS agreed on a value of 100 seconds.  
> > 
> > In fact, that section does not describe how the value is negotiated
> > (since it is not clear what happens if the MS dislikes the value
> > proposed by the AS).  However, this is correct:
> > 
> >    In example in the previous section, the AS and the MS agreed on a
> >    value of 100 seconds.
> 
> [LM] According to RFC6230, the 200 to the SYNC MUST have the same
> keep-alive value proposed in the original SYNC. Who sends the SYNC can
> change according to the COMEDIA negotiation. Specifically, "The previous
> steps ensure that the entity initiating the Control Channel connection
> is always the one specifying the keep-alive timeout period.". So, to
> answer your question, I guess that if a MS dislikes the value, it can
> just pick the active role and send a new SYNC with a new keep-alive
> value.

OK, that was my mistake, as I was confusing how the MRB mechanism
works with the more complicated SIP timeout mechanism.  But I think it
could be clarified by changing the two sentences to:

    In the previous section it has been described that the timeout
    value for the keep-alive mechanism is set by the SYNC request.
    Specifically, in the example the AS specified a value of 100
    seconds.

In particular, the timeout value is not *negotiated* between the AS
and MS, it is simply specified by whichever endpoint takes the active
role.

> > item 22) section 6.1.2
> > 
> >    The 202+REPORT(s) mechanism is used whenever the MS wants to tell
> >    the AS that the requested operation might take some more time than
> >    expected.
> 
> [LM] Actually, the time is Control Package dependent. Different packages
> may have different timeouts. Quoting the framework, "A Control Package
> MUST explicitly define the circumstances under which the server sends
> 200 and 202 messages."

Then the situation could be made clearer by writing:

    The 202+REPORT(s) mechanism is used whenever the MS wants to tell
    the AS that the requested operation might take more time than
    the limit specified in the definition of the Control Package.

> > item 23) section 6.2 figure 18
> > 
> > It might be useful to distinguish the four(?) SIP dialogs involved.
> 
> [LM] Do you mean as in different figures? or clearer text?

I made a mistake there; I should have said "figure 17".

The figure is a bit confusing, as there are either 3 or 4 SIP dialogs
involved, and the SIP messages are not marked to show which dialog
they are within.  (Depending on how the AS functions, the SIP from
UAC2 to AS may or may not be within the same dialog as the SIP from
UAC1 to AS.)  Someone familiar with AS/MS operation can guess which
messages are part of which dialogs, but it may be confusing for those
who are not familiar.  It might not be important in this context, but
in RFCs that come out of the SIP WGs, in a situation like this, the
different SIP dialogs are distinguished by specifying Call-Id values.
In this case, figure 17 would look like this:

   UAC(1)        UAC(2)                  AS                          MS
     |             |                     |                           |
     | INVITE (offer A)                  |                           |
     | Call-Id: A  |                     |                           |
     |---------------------------------->|                           |
     |             |          100 Trying |                           |
     |             |          Call-Id: A |                           |
     |<----------------------------------|                           |
     |             |   INVITE (no offer) |                           |
     |             |   Call-Id: B        |                           |
     |             |<--------------------|                           |
     |             | 180 Ringing         |                           |
     |             | Call-Id: B          |                           |
     |             |-------------------->|                           |
     |             |         180 Ringing |                           |
     |             |          Call-Id: A |                           |
     |<----------------------------------|                           |
     |             |                     | INVITE (offer A)          |
     |             |                     | Call-Id: C                |
     |             |                     |-------------------------->|
     |             |                     |         200 OK (offer A') |
     |             |                     |         Call-Id: C        |
     |             |                     |<--------------------------|
     |             |                     | ACK                       |
     |             |                     | Call-Id: C                |
     |             |                     |-------------------------->|
     |             | 200 OK (offer B)    |                           |
     |             | Call-Id: B          |                           |
     |             |-------------------->|                           |
     |             |                     | INVITE (offer B)          |
     |             |                     | Call-Id: D                |
     |             |                     |-------------------------->|
     |             |                     |         200 OK (offer B') |
     |             |                     |         Call-Id: D        |
     |             |                     |<--------------------------|
     |             |                     | ACK                       |
     |             |                     | Call-Id: D                |
     |             |                     |-------------------------->|
     |             |      ACK (offer B') |                           |
     |             |      Call-Id: B     |                           |
     |             |<--------------------|                           |
     |             |   200 OK (offer A') |                           |
     |             |   Call-Id: A        |                           |
     |<----------------------------------|                           |
     | ACK         |                     |                           |
     | Call-Id: A  |                     |                           |
     |---------------------------------->|                           |
     |             |                     |                           |
     .             .                     .                           .
     .             .                     .                           .

(If the AS acts like a proxy between UAC1 and UAC2 rather than a
B2BUA, then "Call-Id: B" is replaced with "Call-Id: A" throughout.)

> > item 35) section 6.4
> > 
> > The discussion of mixing in the background music should be contrasted
> > with the mixing of the two recordings during playback in section
> > 6.2.3.  Naively, it would seem that the same mixing techniques could
> > be used in both cases, but the two sections implement mixing in
> > entirely different ways.
> 
> [LM] Are you talking of 6.4.2 (current time example)? If so, this was
> done on purpose, to exploit two different scenarios to showcase the use
> of two different techniques. You're right in saying that the same could
> be applied to both cases, nothing naive in this :-)

OK, I see the point of the example.

But I see another issue:  It might help to add to section 3:

   VCR controls:  runtime control of aspects of an audio playback like
      speed and volume, via DTMF signals sent by the user, in a manner that
      resembles the functions of a VCR (video cassette recorder) controller.

> > item 36) section 6.4.3
> > 
> > An example contains XML elements that seem to be unnecessarily
> > complex:
> > 
> >        <rule id="action" scope="public">
> >          <item>
> >            *
> >            <item><ruleref uri="#digit"/></item>
> >          </item>
> >        </rule>
> > 
> > As far as I can tell from the SRGS specification, <item> functions
> > more-or-less as a type of parenthesis, and is needed only when (1) a
> > "weight" or "repeat" attribute needs to be applied to the group, (2)
> > delimiting one alternative of a <one-of> element, or (3) constructing
> > the "null" regular expression (that consumes no input and always
> > matches).  Thus, these examples can be simplified to:
> > 
> >        <rule id="action" scope="public">
> >          *
> >          <ruleref uri="#digit"/>
> >        </rule>
> > 
> > Then again, if the example is taken from a real interaction, perhaps
> > it is best to leave it as it really happened.
> 
> [LM] I'm not an expert of SRGS, so I'm not sure about this. This is how
> we implemented it in our Media Server, and it works this way, but it
> would be nice to hear from other implementers to see if they usually
> realize similar scenarios in a different way!

Yes, let us ask implementers what they think is clearer.

> > item 38) section 6.4.3
> > 
> > This section leads to a question about the semantics of the <volume
> > controltype="setgain" ...> operation:  Clearly, initially the stream
> > is output at the same volume as the input media.  But when "<volume
> > controltype='setgain' value='+1'>" is applied, does that cause the
> > stream's output volume to be set to +1 dB *relative to the input
> > media* or to +1 dB *relative to the previous volume setting*?
> > 
> > E.g., if I apply that operation twice, does the second application
> > leave the amplification at +1 dB or raise it to +2 dB?
> > 
> > The examples in section 6.4.3 suggest the second interpretation.
> > 
> > The text in draft-ietf-mediactrl-mixer-control-package-14 is not clear
> > on this point: "setgain" is only discussed in section 4.2.2.5.1, and
> > it says "use the value of the "value" attribute as a specific gain
> > measured in dB to apply".  That would be most likely to mean "as a
> > specific gain measured in dB to apply to the input", which suggests
> > the first interpretation.
> > 
> > In any case, the semantics of this need to be clarified and the
> > example adjusted accordingly.
> 
> [LM] If I recall correctly, when we first implemented this we discussed
> it with the IVR authors during a meeting. If I'm not wrong the
> conclusion was that setgain *sets* the gain, and does not modify the
> previous value. I don't have the MS code in front of me right now,
> though, I'll have to double check how we implemented this and let you
> know.
> 
> IVR authors, any insight on this from a specification point of view?

Now that I think about it more, I think the usual terminology is that
the gain is seen as the difference between the input and the output,
and "setting" it means fixing the gain at a specific number.  However,
in many usages, a number prefixed by "+" or "-" is used to mean an
increment/decrement relative to the previous value.  What misleads me
is that the setgain operations are shown as implementations of the *1
and *9 requests, which are *incremental*.  The possibility of
misunderstanding could be removed by adding a comment.

(Also, I think that each bullet point should be shown as sentences,
that is, starting with a capital letter and ending with a period,
rather than many clauses separated by ";".  Here, I have left the
paragraphs un-filled to make my change clearer; you probably want to
fill the entire bullet point.)

   o  The AS ACKs the event (B2), and in its business logic understands
      that the code '*1' means that the UAC wants its own volume to be
      lowered in the conference mix.  The AS is able to associate the
      event with the right UAC by referring to the attached dialogid
      (01d1b38).  It then acts accordingly, by sending a <modifyjoin>
      (C1) which does exactly this: the provided <stream> child element
      instructs the MS to modify the volume of the UAC-->conference
      audio flow (setgain=-5dB sendonly).
      Notice that the "setgain" value is the absolute volume level; if
      the user's request is to lower the volume level, the AS must
      remember the previously set volume level and from it calculate
      the new volume level.
      Notice how the request also
      includes directives upon the inverse direction.  This verbose
      approach is needed, since otherwise the MS would not only change
      the volume in the requested direction, but also disable the media
      flow in the other direction; having a proper <stream> addressing
      the UAC<--conf media flow as well ensures that this doesn't
      happen.

Since we have clarified the situation with the first "setgain", there
is no need to repeat the clarification for the second "setgain".

> > item 41) section 7.1, section 7.2, section 7.2.1
> > [...]
> > Generally, "AS" and "MS" are used as the plurals of "AS" and "MS", but
> > in these instances, "ASs" and "MSs" are used.
> 
> [LM] [...] Besides, ASs sounds really bad... :-)

Ah, yes, I overlooked that.  :-)

> > item 46) section 7.2.2.1, section 7.2.2.2
> > 
> >    Please note that, to ease the reading of the protocol contents, a
> >    simple '=_Part' is used whenever a boundary for a 'multipart/mixed'
> >    payload is provided, instead of the actual boundary that would be
> >    inserted in the SIP messages.
> > 
> > This seems like a poor strategy, as the resulting figures are not
> > strictly correct.  Perhaps use "Part" as the boundary value, which
> > would result in figures that look like the following, which strictly
> > correct and are just as easy to read:
> > 
> >    [..]
> >    Content-Type: multipart/mixed;boundary="Part"
> > 
> >    --Part
> >    Content-Type: application/sdp
> > 
> >    v=0
> >    o=- 2890844526 2890842807 IN IP4 as.example.com
> >    [...]
> > 
> >    --Part
> >    Content-Type: application/mrb-consumer+xml
> > 
> >    <?xml version="1.0" encoding="UTF-8" standalone="yes"?>
> >    <mrbconsumer version="1.0"
> >                 xmlns="urn:ietf:params:xml:ns:mrb-consumer">
> >       [...]
> >    </mrbconsumer>
> > 
> >    --Part--
> 
> [LM] I think we did it this way after a comment received on the list:
> Chris and I did the same for the examples in the MRB draft, for instance.
> If you think it might improve the readability of the document we'll
> change this.

I can't find the message that you mention, although the change is
clear between draft-ietf-mediactrl-mrb-03 and -04.  For example:

   1. AS -> MRB (INVITE multipart/mixed)
   -------------------------------------
      [..]
      Content-Type: multipart/mixed; \
			     boundary="----=_Part_0_19138361.1261662356915"

      ------=_Part_0_19138361.1261662356915
      Content-Type: application/sdp

      v=0

becomes

   1. AS -> MRB (INVITE multipart/mixed)
   -------------------------------------
      [..]
      Content-Type: multipart/mixed;boundary="=_Part"

      =_Part
      Content-Type: application/sdp

      v=0

But you could make the example both easy to read and formally correct
by just using "Part" as the boundary value (because no body line
starts with "--Part"):

   1. AS -> MRB (INVITE multipart/mixed)
   -------------------------------------
      [..]
      Content-Type: multipart/mixed;boundary="Part"

      --Part
      Content-Type: application/sdp

      v=0

> > item 51) section 7.3.1
> > 
> > We might want to mention that in this scenario, the AS obtains
> > assignment of an MS *before* any UAC makes a request, in order to have
> > facilities to satisfy future SIP requests.  That is probably how AS's
> > operate in practice, but a naive reader might expect that the AS will
> > make the consumer request only after the AS receives a demand for its
> > services.
> 
> [LM] Actually both Fig.53 and 54 show the AS creating a control channel
> before receiving an INVITE from the UAC. Or was your point about a
> different aspect?

I think my point was that you might want to call out in the text that
the AS creates a control channel to a chosen MS *before* it has any
requests to service.

> > item 53) section 7.3.2
> > 
> >    While apparently not a problem, this raises an issue when the same
> >    unaware AS has several sessions with different MS: the AS would only
> >    see one "virtual" MS (the MRB) and so it would relay all calls there,
> >    making it hard for the MRB to understand where these media dialogs
> >    should belong: specifically, whether the UAC calling belongs to the
> >    AS application logic leading to MS1 or MS2, or wherever else.
> > 
> > This seems to be a deficiency in the CFW protocol:  If the AS
> > establishes a CFW channel with the MS (via a SIP URI to an MRB), and
> > then later the AS wants to addresses a SIP INVITE to the MS, it has no
> > choice but to use the same SIP URI (which routes to the MRB).  This
> > makes it impossible for the MRB to associate the SIP INVITE with the
> > CFW connection, and thus it is difficult for the MRB to route the
> > INVITE to the same MS.
> > 
> > What seems to be needed is for either the COMEDIA response
> > (application/cfw) or CFW connection to provide a SIP URI that is the
> > "real" SIP URI of the far-end of the CFW channel.
> > 
> > Actually, this seems to be a problem even if an AS is addressing an MS
> > URI, but the URI resolves to a redundant set of MSs:  There seems to
> > be no way to ensure that when a CFW channel is established that a
> > subsequent SIP INVITE can be sent to the same MS.  This issue seems to
> > be discussed in RFC 6230 section 3, but I can't find any solution
> > there.
> 
> [LM] The way it works now is that, for a specific MS, the same SIP URI
> is used for both negotiating a control channel and attaching media
> dialogs, yes. A way to overcome this may having the MS redirect media
> dialogs to a different SIP URI when it receives them (that is, treat
> control and media dialogs differently) but I realize how this sounds
> like an implementation hack rather than a real solution.

If an AS is contacting an MS directly, then there is not much of a
problem with the AS using the same URI for negotiating CFW and for
establishing SIP dialogs.  But if the AS is contacting an MRB and the
MRB is contacting an MS, when the MRB receives the SIP INVITE, it must
be able to determine that the AS wants to have the INVITE forwarded to
the same MS to which the MRB forwarded the CFW previously.  How does
the MRB determine that the AS intends the INVITE to go to the same
destination as a particular CFW channel?

> > item 55) section 8
> > 
> > The detailed discussion of how the MS enforces security against the
> > ASs is helpful, but it does not mention how the MS discerns which
> > requests come from AS1 and which come from AS2.

I did not see a response to this issue.

----------------------------------------------------------------------

Summary of uses of "comedia" in published RFCs

RFC 4145 does not contain the term "comedia" at all, although its
short title is "Connection-Oriented Media".

RFC 4571 uses "comedia" only as the name of a Normative Reference to
RFC 4145.

RFC 4572 uses "comedia" only in the short title, "Comedia over TLS in
SDP".

RFC 4975 uses "comedia" only in this sentence:

   The mechanism was loosely based on the Connection-Oriented Media
   (COMEDIA) [26] work done by the MMUSIC working group.

Reference [26] is RFC 4145.

RFC 5567 uses "comedia" only in this sentence:

   [RFC4583] provides a way for SIP UAs to set up a BFCP connection
   towards the Floor Control Server and exploit floor control by means
   of a Connection-Oriented Media (COMEDIA) [RFC4145] negotiation.

RFC 5631 uses "comedia" only in these two sentences:

   This problem could be solved through the use of Connection-Oriented
   Media (COMEDIA) [26], which specifies the "setup" SDP attribute to
   negotiate these roles.

   Assuming it agreed to such a session, it responds with a "setup"
   attribute, as per the COMEDIA specifications.

Reference [26] is RFC 4145.

RFC 6135 uses "comedia" in many places, and includes a definition of
it in the abstract:

   [...] this model uses the connection-oriented media (COMEDIA)
   mechanism in order to create the MSRP transport connection.  The
   model allows MSRP UAs behind Network

RFC 6193 uses "comedia" only within the term "comedia-tls", which is
defined as:

   [...]  "Connection-Oriented Media Transport over the Transport
   Layer Security (TLS) Protocol in the Session Description Protocol
   (SDP)" [RFC4572] (hereafter referred to as comedia-tls) [...]

RFC 6230 uses "comedia" to refer to RFC 4145, and to the SDP attribute
"a=setup".

   The Connection-Oriented Media (COMEDIA) [RFC4145] specification for
   setting up and maintaining reliable connections is used as part of
   the negotiation mechanism (more detail available in later
   sections).

   If the Client initiating the SDP offer has a COMEDIA 'setup'
   attribute equal to active, [...]

RFC 6787 uses "comedia" to refer to RFC 4572:

   When using TLS, the SDP "m=" line for the control stream MUST
   conform to Connection-Oriented Media (COMEDIA) over TLS [RFC4572],
   which specifies the usage of SDP for establishing a secure
   connection-oriented transport over TLS.

RFC 6917 uses "comedia" in two places to refer to SDP support for
connection-oriented media.

----------------------------------------------------------------------

Dale