Re: [rtcweb] Prioritization

[Harald Alvestrand <[e-mail]>]

On 04/25/2014 02:53 PM, Simon Perreault wrote:
> Le 2014-04-25 04:50, Harald Alvestrand a écrit :
>> A.4.  Changes from -03 to -04
>>
>>     o  Added a section on prioritization, moved the DSCP section into it,
>>        and added a section on local prioritization, giving a specific
>>        algorithm for interpreting "priority" in local prioritization.
> Harald,
>
> The new text looks good. A comment about this part:
>
>> 			When an RTCWEB implementation has packets to send on multiple streams	
>> 			that are congestion-controlled under the same congestion controller,	
>> 			the RTCWEB implementation SHOULD serve the streams in a weighted	
>> 			round-robin fashion, with each stream at each level of priority being	
>> 			given approximately twice the transmission capacity (measured in	
>> 			payload bytes) of the level below.
> This looks like a QoS algorithm that might get quickly obsoleted by
> better ones. We don't want to prevent innovation in this matter by
> prescribing an obsolete QoS algorithm.
>
> Suggestion: leave it up to implementations to interpret priority levels
> however they want. Reword the current text so that it becomes an
> *example* of what an implementation might do.

The problem with doing that is that it leads to completely inconsistent 
behaviour.

The reason I wrote it this way is because the bandwidth estimation 
problem breaks down into two parts:

- Telling if it's a good idea to send a packet (congestion control)
- Deciding which packet to send (prioritization)

Congestion control is an active area of research and innovation. I said 
very little about it.

Prioritization is a different matter.

What I was trying to encapsulate was:

- Higher priority means more packets get sent
- "More" has a numeric value, so behaviour is predictable (that's where 
the x2 figure came from)
- Small packet flows get more packets than large-packet flows (counting 
bytes not packets)
- Lower priority flows can't be starved completely

This text captures that. I'd like to continue to capture that.