[Galene] Future plans: Pion v4 and TWCC

[Galene] Future plans: Pion v4 and TWCC

[Juliusz Chroboczek]

Hi,

The reason I've released 0.96 is that it looks like a rock solid version,
and I'm planning to break things.  More specifically, I've just finished
doing a straightforward port to Pion WebRTC v4, and I'm planning to make
the following changes to Galene's protocol:

  - use a separate RTX track for retransmissions;
  - use TWCC instead of REMB for GCC congestion control.

The RTX track will allow browsers to do more precise accounting of packet
loss, which should improve behaviour under packet loss.  Replacing our
home-grown implementation of REMB with Pion's implementation of TWCC will
hopefully improve behaviour on congested links (it will no longer be
necessary to manually switch to "Send: low" when on a slow link), but more
experimentation is needed.

This work is going on in the "pion-v4" branch, which I'll merge into master
when I'm satisfied it doesn't break anything.

-- Juliusz

[Galene] Re: Future plans: Pion v4 and TWCC

[Dave Taht]

On Wed, Jan 1, 2025 at 12:45 PM Juliusz Chroboczek <jch@irif.fr> wrote:
>
> Hi,
>
> The reason I've released 0.96 is that it looks like a rock solid version,
> and I'm planning to break things.  More specifically, I've just finished
> doing a straightforward port to Pion WebRTC v4, and I'm planning to make
> the following changes to Galene's protocol:
>
>   - use a separate RTX track for retransmissions;
>   - use TWCC instead of REMB for GCC congestion control.
>
> The RTX track will allow browsers to do more precise accounting of packet
> loss, which should improve behaviour under packet loss.  Replacing our
> home-grown implementation of REMB with Pion's implementation of TWCC will
> hopefully improve behaviour on congested links (it will no longer be
> necessary to manually switch to "Send: low" when on a slow link), but more
> experimentation is needed.
>
> This work is going on in the "pion-v4" branch, which I'll merge into master
> when I'm satisfied it doesn't break anything.

Did delay and not just loss make it into pion?
> -- Juliusz
> _______________________________________________
> Galene mailing list -- galene@lists.galene.org
> To unsubscribe send an email to galene-leave@lists.galene.org



-- 
Dave Täht CSO, LibreQos

[Galene] Re: Future plans: Pion v4 and TWCC

[Sean DuBois]

[-- Attachment #1: Type: text/plain, Size: 1975 bytes --]

It does [0]

It can use improvements though. I have been distracted by broadcasting/robotics recently, but I would
love to come back to this.

[0] https://github.com/pion/interceptor/blob/master/pkg/gcc/delay_based_bwe.go#L50


> On Jan 1, 2025, at 5:27 PM, Dave Taht <dave.taht@gmail.com> wrote:
> 
> On Wed, Jan 1, 2025 at 12:45 PM Juliusz Chroboczek <jch@irif.fr <mailto:jch@irif.fr>> wrote:
>> 
>> Hi,
>> 
>> The reason I've released 0.96 is that it looks like a rock solid version,
>> and I'm planning to break things.  More specifically, I've just finished
>> doing a straightforward port to Pion WebRTC v4, and I'm planning to make
>> the following changes to Galene's protocol:
>> 
>>  - use a separate RTX track for retransmissions;
>>  - use TWCC instead of REMB for GCC congestion control.
>> 
>> The RTX track will allow browsers to do more precise accounting of packet
>> loss, which should improve behaviour under packet loss.  Replacing our
>> home-grown implementation of REMB with Pion's implementation of TWCC will
>> hopefully improve behaviour on congested links (it will no longer be
>> necessary to manually switch to "Send: low" when on a slow link), but more
>> experimentation is needed.
>> 
>> This work is going on in the "pion-v4" branch, which I'll merge into master
>> when I'm satisfied it doesn't break anything.
> 
> Did delay and not just loss make it into pion?
>> -- Juliusz
>> _______________________________________________
>> Galene mailing list -- galene@lists.galene.org <mailto:galene@lists.galene.org>
>> To unsubscribe send an email to galene-leave@lists.galene.org <mailto:galene-leave@lists.galene.org>
> 
> 
> 
> -- 
> Dave Täht CSO, LibreQos
> _______________________________________________
> Galene mailing list -- galene@lists.galene.org <mailto:galene@lists.galene.org>
> To unsubscribe send an email to galene-leave@lists.galene.org <mailto:galene-leave@lists.galene.org>

[-- Attachment #2: Type: text/html, Size: 10501 bytes --]

[Galene] Re: Future plans: Pion v4 and TWCC

[Juliusz Chroboczek]

> Did delay and not just loss make it into pion?

To my surprise, I've found what looks like a fairly complete
implementation of GCC.  It remains to be seen how well it works in
practice.

Dave, interested in collaborating on making some benchmarks?

-- Juliusz

[Galene] Re: Future plans: Pion v4 and TWCC

[Dave Taht]

On Wed, Jan 1, 2025 at 5:16 PM Sean DuBois <sean@pion.ly> wrote:
>
> It does [0]
>
> It can use improvements though. I have been distracted by broadcasting/robotics recently, but I would
> love to come back to this.

yay!

>
> [0] https://github.com/pion/interceptor/blob/master/pkg/gcc/delay_based_bwe.go#L50

sigh. this reacts so slowly.... rc := newRateCalculator(500 * time.Millisecond)

Starlink would be interesting.
>
>
> On Jan 1, 2025, at 5:27 PM, Dave Taht <dave.taht@gmail.com> wrote:
>
> On Wed, Jan 1, 2025 at 12:45 PM Juliusz Chroboczek <jch@irif.fr> wrote:
>
>
> Hi,
>
> The reason I've released 0.96 is that it looks like a rock solid version,
> and I'm planning to break things.  More specifically, I've just finished
> doing a straightforward port to Pion WebRTC v4, and I'm planning to make
> the following changes to Galene's protocol:
>
>  - use a separate RTX track for retransmissions;
>  - use TWCC instead of REMB for GCC congestion control.
>
> The RTX track will allow browsers to do more precise accounting of packet
> loss, which should improve behaviour under packet loss.  Replacing our
> home-grown implementation of REMB with Pion's implementation of TWCC will
> hopefully improve behaviour on congested links (it will no longer be
> necessary to manually switch to "Send: low" when on a slow link), but more
> experimentation is needed.
>
> This work is going on in the "pion-v4" branch, which I'll merge into master
> when I'm satisfied it doesn't break anything.
>
>
> Did delay and not just loss make it into pion?
>
> -- Juliusz
> _______________________________________________
> Galene mailing list -- galene@lists.galene.org
> To unsubscribe send an email to galene-leave@lists.galene.org
>
>
>
>
> --
> Dave Täht CSO, LibreQos
> _______________________________________________
> Galene mailing list -- galene@lists.galene.org
> To unsubscribe send an email to galene-leave@lists.galene.org
>
>


-- 
Dave Täht CSO, LibreQos

[Galene] Re: Future plans: Pion v4 and TWCC

[Dave Taht]

On Wed, Jan 1, 2025 at 5:49 PM Juliusz Chroboczek <jch@irif.fr> wrote:
>
> > Did delay and not just loss make it into pion?
>
> To my surprise, I've found what looks like a fairly complete
> implementation of GCC.  It remains to be seen how well it works in
> practice.
>
> Dave, interested in collaborating on making some benchmarks?

yes!

>
> -- Juliusz



-- 
Dave Täht CSO, LibreQos

[Galene] Re: Future plans: Pion v4 and TWCC

[Tim Panton]

I was talking to a friend about starlink and bwe. He claims the variation is predictable.
If true it might be fun to do an estimator that -knew- about starlink and scheduled a full frame immediately after the handover glitch. (For example)

T.

Sent from my iPad

> On 2 Jan 2025, at 18:57, Dave Taht <dave.taht@gmail.com> wrote:
> 
> On Wed, Jan 1, 2025 at 5:16 PM Sean DuBois <sean@pion.ly> wrote:
>> 
>> It does [0]
>> 
>> It can use improvements though. I have been distracted by broadcasting/robotics recently, but I would
>> love to come back to this.
> 
> yay!
> 
>> 
>> [0] https://github.com/pion/interceptor/blob/master/pkg/gcc/delay_based_bwe.go#L50
> 
> sigh. this reacts so slowly.... rc := newRateCalculator(500 * time.Millisecond)
> 
> Starlink would be interesting.
>> 
>> 
>> On Jan 1, 2025, at 5:27 PM, Dave Taht <dave.taht@gmail.com> wrote:
>> 
>> On Wed, Jan 1, 2025 at 12:45 PM Juliusz Chroboczek <jch@irif.fr> wrote:
>> 
>> 
>> Hi,
>> 
>> The reason I've released 0.96 is that it looks like a rock solid version,
>> and I'm planning to break things.  More specifically, I've just finished
>> doing a straightforward port to Pion WebRTC v4, and I'm planning to make
>> the following changes to Galene's protocol:
>> 
>> - use a separate RTX track for retransmissions;
>> - use TWCC instead of REMB for GCC congestion control.
>> 
>> The RTX track will allow browsers to do more precise accounting of packet
>> loss, which should improve behaviour under packet loss.  Replacing our
>> home-grown implementation of REMB with Pion's implementation of TWCC will
>> hopefully improve behaviour on congested links (it will no longer be
>> necessary to manually switch to "Send: low" when on a slow link), but more
>> experimentation is needed.
>> 
>> This work is going on in the "pion-v4" branch, which I'll merge into master
>> when I'm satisfied it doesn't break anything.
>> 
>> 
>> Did delay and not just loss make it into pion?
>> 
>> -- Juliusz
>> _______________________________________________
>> Galene mailing list -- galene@lists.galene.org
>> To unsubscribe send an email to galene-leave@lists.galene.org
>> 
>> 
>> 
>> 
>> --
>> Dave Täht CSO, LibreQos
>> _______________________________________________
>> Galene mailing list -- galene@lists.galene.org
>> To unsubscribe send an email to galene-leave@lists.galene.org
>> 
>> 
> 
> 
> -- 
> Dave Täht CSO, LibreQos
> _______________________________________________
> Galene mailing list -- galene@lists.galene.org
> To unsubscribe send an email to galene-leave@lists.galene.org