[Galene] Re: using up more ports in ipv6 for better congestion control

[T H Panton <tim@pi.pe>]

> On 16 Jul 2021, at 03:37, Juliusz Chroboczek <jch@irif.fr> wrote:
> 
>>> Tim, I'd be very grateful if you could explain what advantages TWCC has
>>> over REMB.  For now, I'm sticking with REMB.
> 
>> I can’t speak from experience (I’ve only used REMB) - but my sense is
>> that the difference really kicks in when you have multiple video media
>> streams using the same path.  So perhaps video, screen share and
>> audio. REMB treats each stream separately if I recall.
> 
> Then REMB could be modified to perform per-connection congestion control,
> just like TWCC, without all of the chattiness of TWCC.

I don’t think it could. - isn't REMB based on watching the packet arrival interval which is
pretty consistent on a single stream.  But imagine multiplexing some opus (50 fps)
a screenshare (10fps) a thumbnail (15fps) and a presenter view (60 fps), you now have
multiple valid packet intervals (in some sort of repeating pattern). 
I imagine that would make the smoother in REMB go badly wrong.

TWCC is based on watching the time of flight - which doesn’t have that problem.

> 
> I really feel that I'm missing something.
> 
> -- Juliusz