Re: [AVTCORE] [rtcweb] [tsvwg] WG Last Call on changes: draft-ietf-avtcore-rtp-circuit-breakers-16

[Colin Perkins <[e-mail]>]

> On 29 Jun 2016, at 15:54, Michael Welzl <[e-mail]> wrote:
>> On 29. jun. 2016, at 16.44, Black, David <[e-mail]> wrote:
>> I wrote: 
>> 
>>>> Another possible rationale for this mixing is that if drops start occurring, then many of
>>>> the new and proposed uses of ECN that treat ECN-CE marks as less than loss-equivalent
>>>> are outside their intended operating envelopes/regions.
>> 
>> Colin responded:
>> 
>>> Clearly if the queue has been driven to overflow, so that packet loss is
>>> occurring, then the AQM is outside its intended operating regime. I’m not sure
>>> we need to push it so far, though. Is there not a regime where the ECN-CE
>>> marking rate indicates excessive congestion, before the queue overflows and
>>> drops packets?
>> 
>> Yes, but that may not be relevant to this discussion.   My hypothesis for this discussion
>> is that actual drops will occur well before RTP  is running at 10x the TCP equivalent rate
>> (based on drops and ECN-CE marks), and that 10x factor is the trip threshold for the
>> circuit breaker, which is the focus of this discussion. 
>> 
>> I would think that the "regime where the ECN-CE marking rate indicates excessive
>> congestion, before the queue overflows and drops packets?" would not extend as
>> far as RTP running at 10x the TCP equivalent rate, which is where the RTP circuit
>> breaker trips.
> 
> Seems reasonable to me….

Does anyone have data to let us find out either way? 

>> This is all a thought exercise - I'm happy to be shown to be wrong based on actual
>> data and experience with "running code" ...
>> 
>> To which Michael responded: 
>> 
>>> Shouldn’t a congestion control mechanism react well before that?
>> 
>> Yes, if there is one ;-).  This is RTP …
> 
> Earlier, in the same conversation, so - I assume - the same context, Colin wrote:
> "Assuming, in all cases, that there’s a parallel congestion control algorithm running”

If the congestion control is working, then it should adapt the rate so the circuit breaker never triggers. Yes, there should be a parallel congestion control, but for the circuit breaker to be relevant, that congestion control is either not working properly, or not implemented. 

-- 
Colin Perkins
https://csperkins.org/