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

[Sean DuBois <sean@siobud.com>]

On Sun, Jul 11, 2021 at 01:19:47PM +0200, Juliusz Chroboczek wrote:
> Tim is right about everything, as usual.
> 
> Tim said:
> 
> >> "It is still possible. Just set bundle-policy to max-compat and you'll
> >> get one stream for audio and one for video.
> 
> This mode of operation is not currently supported by Pion.
> 
> Tim also said:
> 
> > Oh, no, the congestion control is _very_ much alive in webRTC but under
> > guise of bandwidth estimation - The ‘simplest' is google’s REMB RTCP
> > message which basically looks at the arrival time of packets and uses
> > any _lengthening_ in the tof to deduce the onset of additional buffering
> > in the path.
> 
> Yes.  Galène implements both REMB and loss-based congestion control in the
> sender->SFU direction, and just loss-based in the SFU->receiver direction.
> Implementing REMB is on my to-do list, but I've got more important things
> to do.
> 
> Dave said:
> 
> >> I am also under the impression that the congestion control
> >> notifications in rtcp are essentially obsolete in the rfc, mandating a
> >> 500ms interval instead of something sane, like a frame?
> 
> You're speaking about the loss-based congestion controller, which is used
> in combination with the delay-based controller that Tim is referring to.
> The 500ms interval is the default, but nothing prevents the receiver from
> sending more frequent feedback, for example just after a loss episode.
> (Galène doesn't currently do that.)
> 
> Tim said:
> 
> > Transport CC tries to expand this to apply to all the streams muxed over
> > a port by adding an RTP header extension with an accurate (NTP) clock in
> > it.
> 
> I still don't understand why Transport-wide CC (TWCC) is better than REMB.
> In my view, it's just moving the REMB congestion controller from the
> receiver to the sender, which requires huge amounts of timely per-packet
> feedback.
> 
> Tim, I'd be very grateful if you could explain what advantages TWCC has
> over REMB.  For now, I'm sticking with REMB.
> 

TWCC gives you the interarrival time of packets. You also get this with
abs-send-time and do REMB. I would be interested to know the math of
data costs of abs-send-time vs TWCC (reference time + deltas)

With TWCC the sender knows the metadata of lost packets. If you lose a packet
with REMB you don't know the send time or the size of the packet. That
seems like it could be useful information?

> > I realise this is anathema to TCP folks, it certainly came as a shock to me
> 
> It depends on your background, I guess.  It was fairly natural for me, but
> then I've been brought up on the litterature on ECN on the one hand and
> TCP-Vegas on the other, both of which aim to perform congestion control
> without any packet drops.
> 
> Ceterum autem censeo that Dave is right, and that we should be working on
> ECN support in WebRTC.

Getting things into Chromium has been hard for me. If ECN gets
acceptance that would be amazing. My mindset is that TWCC is most
effort/least reward.

Pion just got a Network Conditioner so I am hoping to put a basic TWCC
driven congestion controller in pion/transport. A few papers were
published on Google Congestion Control. The code [0] is nicely split up
though (Delay, Loss based BWE all have their own classes)

> 
> -- Juliusz

[0] https://source.chromium.org/chromium/chromium/src/+/main:third_party/webrtc/modules/congestion_controller/goog_cc/