Re: [rmcat] Priority and rtcweb

[Harald Alvestrand <[e-mail]>]

On 11/13/2017 09:44 AM, Michael Welzl wrote:
> Can I hijack this thread to talk about priority?
>
> There was a thread about priority in rtcweb - someone asking for a
> float… coupled-cc can easily support a float, but Harald’s transport
> draft doesn’t support it. Opinions went back and forth for a bit...
> I think we should get more WebRTC deployment and not more debate on
> small details that delays the work, so I don’t want to be a
> troublemaker and didn’t interfere with this discussion.
> But, what is the view of people in rmcat about this?

the reason for the 4 levels was indeed that they were defined at a time
when we thought that any underlying system (whether it be diffserv or
congestion controllers that supported float priorities) could do
something sensible with them, while having more than 4 things to choose
from seemed to be of questionable value in a large majority of use cases.

a float doesn't help much if you don't know what the effect of changing
the value is; is it linear in bandwidth, linear in bandwidth fraction 
(sum and normalize? 0..1?), exponential in some dimension, logarithmic
in some other dimension?

If a congestion controller mechanism wants to interface to the webrtc 4
levels, it would have to do something like the diffserv-to-priority
document from TSV - defining what controls on the congestion controller
one needs to set to have an appropriate result on what gets sent when
congestion happens.




>
> Cheers,
> Michael
>
>
>> On Nov 13, 2017, at 2:41 PM, Varun Singh <[e-mail]
>> <mailto:[e-mail]>> wrote:
>>
>> Hi,
>>
>> Was there a study done for the parameters needed for congestion
>> control? I know we debated this at length, were there slides created
>> by  the congestion control proponents, that I can reference. 
>>
>> From the top of my head, I remember there’s 
>> + priority (needed by coupled congestion control)
>>
>> I’m asking this for w3c input. They already have max bitrate, max
>> framearte, and degradation preference  degradation preference
>> indicates where the bits should be allocated: in maintaining frame
>> rates or frame sizes. 
>>
>> Feedback is appreciated To make sure we didn’t miss anything.
>>
>> Cheers,
>> Varun. 
>> -- 
>> Founder, CEO, callstats.io <http://callstats.io/>
>> http://www.callstats.io <http://www.callstats.io/>
>>
>> Interested in networking, media quality, and diagnostics.
>> We are hiring!: www.callstats.io/jobs/ <http://www.callstats.io/jobs/>
>

-- 
Surveillance is pervasive. Go Dark.