Re: [MMUSIC] BUNDLE and RTCP

IMHO having a separate port for RTCP is such a pain in general and with ICE in particular that we really should just mandate RTP/RTCP muxing for any new RFCs.  In particular anything that legacy equipment won't ever support anyway.  As a separate port for RTCP does involve considerable additional complexity and cost in development and testing to support, there has to be tangible ROI for it as a feature.  I'd say those who want to keep the separate port (be it RTP+1 or through a=rtcp) need to justify why the added complexity is worth it.  I also agree with Paul that the separate port for RTCP was a mistake in the original design--at least in hindsight :-)

Thanks,
--Felix

________________________________________
From: mmusic <mmusic-bounces@ietf.org> on behalf of Christer Holmberg <christer.holmberg@ericsson.com>
Sent: Monday, October 19, 2015 13:48
To: Paul Kyzivat; mmusic@ietf.org
Subject: Re: [MMUSIC] BUNDLE and RTCP

Hi,

>>> Good discussion.
>>>
>>> How about the following:
>>>
>>> If there is an RTP m-line in the bundle, and rtcp-mux is *not*
>>> negotiated, then the rtcp port is determined according to the rules
>>> for the first m-line selected in the bundle group. If *that* line has, and allows, an a=rtcp, then that is used. If that line doesn't have a=rtcp, then port+1 is used, even if this isn't an rtp m-line.
>>
>> Sure. But, as the offerer doesn't know for sure which m- line will be selected, it always has to be prepared to use port+1, so why not use port+1 to begin with?
>>
>> I guess my question is whether someone really needs a=rtcp with BUNDLE? As far as I know, most current implementations do rtcp-mux, and for those that don't/can't there is always port+1.
>
> IIRC the original reason for a=rtcp was because there are cases when it is hard to get an even/odd pair of ports. If the offerer is in that situation, then mandating port+1 isn't going to work.

Ok. So, I guess one question is whether such situations still can occur.

> I am coming around to agree with just requiring rtcp-mux with bundle.
> I don't unsderstand why RTCP was *ever* defined to run on a separate port. Maybe it was just a mistake.

As far as I know, some people still wish to be able to use separate ports for RTP and RTCP, even when using BUNDLE. I'd like to know whether port+1 works for them.

Regards,

Christer

>> (Right now there isn't any way to have a bundle group with more than
>> one non-rtp m-line, but that might not always be so.)
>
> Correct.
>
> Regards,
>
> Christer
>
>
>
> On 10/19/15 3:35 AM, Christer Holmberg wrote:
>> Hi,
>>
>> ...
>>
>> ...
>>
>> ...
>>
>>      HOWEVER, we could make it SIMPLE and either:
>>
>>      1)*Mandate usage of rtcp-mux with BUNDLE*. I.e. if BUNDLE is
>>      negotiated, rtcp-mux MUST be used.
>>
>>      This was already suggested in the past, but Paul(?) said we should
>>      not make such restriction without a good reason. I think the current
>>      issue is a good reason :)
>>
>>      m=data channel 10000
>>
>>      m=rtp 11111
>>
>>      a=rtcp-mux
>>
>>      m=rtp 11222
>>
>>      a=rtcp-mux
>>
>>      2)*Mandate usage of either rtcp-mux OR the default "+1" port with
>>      BUNDLE*. I.e. if BUNDLE is negotiated, rtcp-mux or "+1" MUST be
>>      used. The selection is based on whether the rtcp-mux attribute was
>>      included in the offer/answer or not.
>>
>>      m=data channel 10000             // rtcp-mux
>>
>>      m=rtp 11111
>>
>>      a=rtcp-mux
>>
>>      m=rtp 11222
>>
>>      a=rtcp-mux
>>
>>      m=data channel 10000             // "+1"
>>
>>      m=rtp 11111
>>
>>      m=rtp 11222
>>
>>      The solutions above would not allow explicit negotiation of an RTCP
>>      port when BUNDLE is used, but maybe we could live with that?
>>
>>> Well, I could live with both of your suggestions, but I thought you
>> wanted a solution that can also negotiate which a=rtcp attribute to use.
>>
>>> If nobody really needs something like that , I would prefer 1) as it
>> removes the needs for several implementation option that would only
>> be needed to cover full backwards compatibility but would hardly be used.
>>
>> Personally I don't have a strong preference, but as far as I know
>> there may be environments where separate ports for RTP and RTCP are
>> desired - even if BUNDLE is used.
>>
>>>> Now, if we can agree on a way forward before Yokohama, you won't
>>>> have to
>> sit listening to me talking about BUNDLE at the meeting :)
>>>
>>> Well, I wouldn't be sitting there anyhow but understand that your
>> motivation to speak there about BUNDLE again is probably limited.
>>
>> I don't mind talking about BUNDLE - but I'd like to talk about it in
>> the context of the great RFC that we have published :)
>>
>> Regards,
>>
>> Christer
>>
>>
>>
>> _______________________________________________
>> mmusic mailing list
>> mmusic@ietf.org
>> https://www.ietf.org/mailman/listinfo/mmusic
>>
>
> _______________________________________________
> mmusic mailing list
> mmusic@ietf.org
> https://www.ietf.org/mailman/listinfo/mmusic
>

_______________________________________________
mmusic mailing list
mmusic@ietf.org
https://www.ietf.org/mailman/listinfo/mmusic