Re: [Dart] [AVTCORE] Treatment of RTCP (was Re: Colin Perkins comments - WGLC: draft-ietf-dart-dscp-rtp-02)

Hi,

Where are we with this discussion? Are there conclusions buried in there somewhere?

Thanks!

Ben.

On Aug 27, 2014, at 1:50 PM, Colin Perkins <csp@csperkins.org> wrote:

> On 27 Aug 2014, at 19:47, Michael Welzl <michawe@ifi.uio.no> wrote:
>> On 27. aug. 2014, at 20:42, Colin Perkins <csp@csperkins.org> wrote:
>>> On 27 Aug 2014, at 19:39, Michael Welzl <michawe@ifi.uio.no> wrote:
>>>> On 27. aug. 2014, at 18:00, Colin Perkins <csp@csperkins.org> wrote:
>>>>> On 26 Aug 2014, at 17:38, Black, David <david.black@emc.com> wrote:
>>>>>>> The more difficult case is when an SSRC is sending video using different
>>>>>>> markings for RTP packets carrying the I- and P-frames. Should that SSRC then
>>>>>>> mark its RTCP packets like the RTP packets carrying I-frames, like the RTP
>>>>>>> packets carrying P-frames, or what?
>>>>>> 
>>>>>> Answering a question w/a question :-), how are those reports likely to be used?
>>>>>> 
>>>>>> For example, if the primary use of these reports is to adjust a variable rate
>>>>>> codec's sending rate, the P-frame info is probably more useful as indicative
>>>>>> of what's happening to the traffic that the network drops first when the going
>>>>>> gets rough (or whose delivery w/o loss indicates that a sending rate increase
>>>>>> may be reasonable), which suggests P-frame-like RTCP report marking.
>>>>> 
>>>>> I doubt the RTT estimate derived from RTCP is used for congestion control, since it’s too infrequent to get insight into the dynamics. It’s for user experience reporting, maybe rough clustering of receivers, that sort of thing. 
>>>> 
>>>> I'd agree if I didn't have the impression, in RMCAT, that nothing but RTP / RTCP is allowed?!  So can we send extra packets to probe for the RTT?
>>> 
>>> I’d expect RMCAT will probe the one-way delay variation by measuring the arrival times of RTP packets, and report that in RTCP once per video frame (or so). That would give a much more accurate measurement than a single-packet sample of the RTT derived from RTCP, and would still allow per-frame adaptation.
>> 
>> I agree that this is a useful signal to use - but I think you still need a rough idea of the RTT to be able to interoperate with TCP, else you may cause TCP to go into RTOs and be ignorant of that. But this could perhaps be a more coarse signal - maybe that's okay. I think this is reasonable.
> 
> You have the RTT estimate, irrespective of the other information you include along with the regular RTCP packets, although it’s relatively infrequent and coarse-grained.
> 
>> I wonder if your RTCP draft in RMCAT should clarify this?  (sorry if it does already, didn’t check)
> 
> It probably should talk about this, I agree.
> 
> 
> -- 
> Colin Perkins
> http://csperkins.org/
> 
> 
> 
> _______________________________________________
> Dart mailing list
> Dart@ietf.org
> https://www.ietf.org/mailman/listinfo/dart