Re: [rtcweb] [Suspected Junk Mail] Endpoints that don't support RTCP

[Harald Alvestrand <[e-mail]>]

Den 23. april 2015 21:44, skrev Cullen Jennings:
> 
>> On Apr 22, 2015, at 4:44 PM, Colin Perkins <[e-mail]> wrote:
>>
>>> Perpetuating tolerance for the broken behavior of ignoring "MUST send
>>> RTCP" seems like a path that doesn't lead to getting rid of the behavior
>>> in the end.
>>
>> I tend to agree: this seems like something where the gateway ought to handle the interoperability, and synthesise appropriate RTCP. 
>>
>> Or, if that's really too onerous, tie into some other consent check that can indicate that the flow is still wanted and not causing problems.
> 
> People keep talking about gateway's but I'm talking about endpoints that do ICE and DTLS-SRTP but don't do RTCP. They also don't do video. 
> 
> The reason they don't do RTCP is because they don't perceive a value in it. I'm not arguing this is right or wrong, I'm just saying they don't see the value. The point the folks that think this is a good idea make is that if RTCP says there is X jitter, or Y % packet loss, there is nothing they can do with this. They will not stop the call until the audio is so bad that user hangs up. Only the user hanging up stops the call. Thus the only thing RTCP does is cause more congestion. Again, not saying this is right but given that most the audio only RTP I see on the network does not have RTCP, I don't think this is likely to change any time soon.  And I want to say again, for video it is the opposite story, the bulk of the interactive video RTP I see has RTCP. 
> 

And I'm arguing (thinking out loud) that if we condone this behavior by
successfully interoperating with these devices from WebRTC
installations, we're prolonging the problem.

If our consensus (or consensus-minus-Emil) is that we need devices to
back off on sending RTP in the face of severe congestion for the sake of
the network, and our consensus is that we need RTCP in order to detect
that condition, the only reasonable conclusion is that we should refuse
to interoperate with those devices that do not send RTCP.

These devices already upgraded to ICE and DTLS-SRTP, presumably because
they wanted to interoperate with WebRTC installations. Sending an RTCP
packet once every few seconds is not going to be what kills them.