Re: [MMUSIC] bundle-11: BUNDLE accpeted, but RTCP Mux rejected

["Stach, Thomas" <[e-mail]>]

inline

From: Christer Holmberg [mailto:[e-mail]]
Sent: Mittwoch, 8. Oktober 2014 13:37
To: Stach, Thomas; Harald Alvestrand; Cullen Jennings (fluffy)
Cc: mmusic
Subject: RE: bundle-11: BUNDLE accpeted, but RTCP Mux rejected

Hi Thomas,

I suggest something like this:

In section 10.3.2.3:

OLD TEXT:

"If the answerer rejects usage of RTP/RTCP multiplexing within the
                BUNDLE group, it MUST NOT assign an SDP 'rtcp-mux' or SDP 'rtcp'
                attribute to any bundled "m=" line in the answer."

NEW TEXT:

"If the answerer rejects usage of RTP/RTCP multiplexing within the
                BUNDLE group, it MUST NOT assign an SDP 'rtcp-mux' or SDP 'rtcp'
                attribute to any bundled "m=" line in the answer. The answerer will,
                based on the port number of the selected offerer BUNDLE address,
                use the next higher (odd) destination port number [RFC3550] for
sending RTCP packets associated with a bundled "m=" line towards
the offerer,"
[TS] I'd like to add something for receiving RCTP. With a slight re-wording the new text would be
If the answerer rejects usage of RTP/RTCP multiplexing within the
BUNDLE group, it MUST NOT assign an SDP 'rtcp-mux' or SDP 'rtcp'
attribute to any bundled "m=" line in the answer.
 Based on the port number of the selected offerer BUNDLE address,
 the answerer will send RTCP packets associated with any bundled "m=" line
 to the next higher (odd) destination port [RFC3550].
 Based on the port number of the selected answerer BUNDLE address,
 the answerer will receive RTCP packets associated with any bundled "m=" line
 at the next higher (odd) destination port [RFC3550].
This means that the associated RTCP streams for bundled "m=" lines
 are bundled at the next higher (odd) destination port


In section 10.3.2.4:

OLD TEXT:

                "If the answerer does not accept the usage of RTP/RTCP multiplexing
                [Section 10.3.2.3], the offerer MUST use separate 5-tuples for RTP
                and RTCP."

NEW TEXT:

                "If the answerer does not accept the usage of RTP/RTCP multiplexing
                [Section 10.3.2.3], the offerer MUST use separate 5-tuples for RTP
                and RTCP. The answerer will, based on the port number of the answerer
BUNDLE address, use the next higher (odd) destination port number [RFC3550]
for sending RTCP packets associated with a bundled "m=" line towards the
answerer."

[TS] Similar here
If the answerer does not accept the usage of RTP/RTCP multiplexing
[Section 10.3.2.3], the offerer MUST use separate 5-tuples for RTP
and RTCP.
Based on the port number of the selected offerer BUNDLE address,
 the offerer will receive RTCP packets associated with any bundled "m=" line
 at the next higher (odd) destination port [RFC3550].
 Based on the port number of the selected answerer BUNDLE address,
 the offerer will send RTCP packets associated with any bundled "m=" line
 to the next higher (odd) destination port [RFC3550].


Regards,

Christer




From: Stach, Thomas [mailto:[e-mail]]
Sent: 8. lokakuuta 2014 13:06
To: Christer Holmberg; Harald Alvestrand; Cullen Jennings (fluffy)
Cc: mmusic
Subject: RE: bundle-11: BUNDLE accpeted, but RTCP Mux rejected


Christer,

From: Christer Holmberg [mailto:[e-mail]]
Sent: Dienstag, 7. Oktober 2014 20:35
To: Stach, Thomas; Harald Alvestrand; Cullen Jennings (fluffy)
Cc: mmusic
Subject: RE: bundle-11: BUNDLE accpeted, but RTCP Mux rejected

Hi Thomas,

>BUNDLE can be used separate from RFC5761 RTCP-Multiplexing.
>I can't find text in bundle-11 where the answerer would send its RTCP packets if it accepted BUNDLE, but rejected RTCP-muxing.
>I see two options:
>1. The answerer sends all RTCP packets to the shared RTP-port+1 of the negotiated shared address.
>2. The answerer sends the RTCP packets to separate ports based on the unique addresses in the m-lines of the offer?

Alternative 1) is correct. I guess we could add some text to clarify that.
[TS] Yes, please. The current text is not sufficiently clear.

>The offerer on the other hand will only receive a shared address. I assume it then has to send all its RTCP packets
>to the shared RTP-port+1 of the answerer.

Yes.

>Is my understanding correct or am I completely off-track?

Assuming my understanding is correct, your understanding is correct :)

>Supposed I'm correct, then independent from the response, I'm asking if there really is a use case for that.
>In case 1, one would have to demux the RTP streams and RTCP streams from two separate ports instead of a single port. How much implementation effort >is saved in that case?

I am not sure what you mean. Are you asking how much implementation effort is saved if you don't have to de-mux RTP and RTCP?
[TS] Yes.  I was questioning myself what benefit there is in not mandating RTCP-mux.
In the meantime I looked up again the discussion on "Q14" that led to having rtcp-mux optional. I agree with the outcome.
Thanks!

>In case 2 the RTCP handling at offerer and answerer is completely different. In addition, you still need "a lot" of RTCP ports and take advantage of only half >of the BUNDLE benefits.

Correct. But, case 2 is not applicable (unless the endpoints choose not to use BUNDLE to begin with, that is...).

Regards,

Christer