On 12. okt. 2013, at 08:28, Mikael Abrahamsson <[email protected]> wrote:

> On Sat, 12 Oct 2013, Michael Welzl wrote:
> 
>> My question is, why is none of this happening?
> 
> My guess is that the rule to always deliver packets within a stream in order. 
> If packets within the same stream have different DCSP values, there is a risk 
> that they will go into different queues and they might be delivered out of 
> order. A lot of applications that do real-time will have a PDV buffer, but 
> won't have much of a out-of-order buffer. If a packet with sequence number 10 
> arrives after sequence number 8, packet number 9 is often assumed to be lost.
> 
> AQM is not widely deployed at all on the Internet as it is today, and a lot 
> of devices will just have 4 different queues. Also as you say below, putting 
> packets into different queues might mean they compete with other traffic and 
> you might get packet loss due to that.
> 
> It would of course be of interest going forward to look into your proposal. 
> AQM/Bufferbloat movement could perhaps benefit from this, but the current 
> thinking is to make sure that all streams get equal access to the media with 
> minimal influence between them, not that much to decide what packets within a 
> stream to select for drop.

Thanks, that's about what I guessed... strange, a bit, given that it's not hard 
to show quite large quality gains if packet priorities within a stream 
(five-tuple) are honored, and doing per-5-tuple things near the edge doesn't 
seem all too unusual to me...

So indeed it wasn't a proposal, I was just wondering why it's not done. Maybe 
not enough interest from the parties involved (because application developers 
are not usually the same people that build middle-boxes)

Cheers,
Michael

Reply via email to