[Ecn-sane] Multiplicative Decrease Scaling (high-fidelity ECN in a single queue)

[Ecn-sane] Multiplicative Decrease Scaling (high-fidelity ECN in a single queue)

[Pete Heist]

Hi all,

I'm looking for some feedback on a way to do early and frequent ECN
that retains compatibility with conventional CCAs in a single queue.
Here's a write-up including simulation results:

https://github.com/heistp/md-scaling/blob/main/MD-Scaling.md

-----
Abstract:

Multiplicative Decrease Scaling (MDS or MD-Scaling) provides high-
fidelity Explicit Congestion Notification (ECN) in a way that's
compatible with conventional congestion control algorithms (CCAs) in a
single queue. Using a signaling and response relationship that allows
cwnd convergence with conventional CCAs, MDS aims to reduce queueing
delay, and increase utilization for MDS CCAs, while supporting the
transition to scalable CCAs.
-----

I find the results interesting, and am looking to determine if this
idea is worth pursuing further.  Any feedback on the approach, proposed
signaling mechanism or anything else is welcome.

Thanks!
Pete Heist

PS- I've also posted this to ICCRG in the IRTF, so apologies to anyone
getting hit with it twice.

Re: [Ecn-sane] Multiplicative Decrease Scaling (high-fidelity ECN in a single queue)

[Sebastian Moeller]

Hi Pete,

nice! I note that the legend for figure 8 b says: "Figure 8b: CUBIC-MDS at 100 Mbps, 600 ms path RTT" but the plot implies 1000 Mbps.

> On 21. Oct 2024, at 10:48, Pete Heist via Ecn-sane <ecn-sane@lists.bufferbloat.net> wrote:
> 
> Hi all,
> 
> I'm looking for some feedback on a way to do early and frequent ECN
> that retains compatibility with conventional CCAs in a single queue.
> Here's a write-up including simulation results:
> 
> https://github.com/heistp/md-scaling/blob/main/MD-Scaling.md
> 
> -----
> Abstract:
> 
> Multiplicative Decrease Scaling (MDS or MD-Scaling) provides high-
> fidelity Explicit Congestion Notification (ECN) in a way that's
> compatible with conventional congestion control algorithms (CCAs) in a
> single queue. Using a signaling and response relationship that allows
> cwnd convergence with conventional CCAs, MDS aims to reduce queueing
> delay, and increase utilization for MDS CCAs, while supporting the
> transition to scalable CCAs.
> -----
> 
> I find the results interesting, and am looking to determine if this
> idea is worth pursuing further.  Any feedback on the approach, proposed
> signaling mechanism or anything else is welcome.
> 
> Thanks!
> Pete Heist
> 
> PS- I've also posted this to ICCRG in the IRTF, so apologies to anyone
> getting hit with it twice.
> _______________________________________________
> Ecn-sane mailing list
> Ecn-sane@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/ecn-sane

Re: [Ecn-sane] Multiplicative Decrease Scaling (high-fidelity ECN in a single queue)

[Pete Heist]

Thanks Sebastian, fixed that typo...

Pete

On Mon, 2024-10-21 at 23:46 +0200, Sebastian Moeller wrote:
> Hi Pete,
> 
> nice! I note that the legend for figure 8 b says: "Figure 8b: CUBIC-
> MDS at 100 Mbps, 600 ms path RTT" but the plot implies 1000 Mbps.
> 
> > On 21. Oct 2024, at 10:48, Pete Heist via Ecn-sane
> > <ecn-sane@lists.bufferbloat.net> wrote:
> > 
> > Hi all,
> > 
> > I'm looking for some feedback on a way to do early and frequent ECN
> > that retains compatibility with conventional CCAs in a single
> > queue.
> > Here's a write-up including simulation results:
> > 
> > https://github.com/heistp/md-scaling/blob/main/MD-Scaling.md
> > 
> > -----
> > Abstract:
> > 
> > Multiplicative Decrease Scaling (MDS or MD-Scaling) provides high-
> > fidelity Explicit Congestion Notification (ECN) in a way that's
> > compatible with conventional congestion control algorithms (CCAs)
> > in a
> > single queue. Using a signaling and response relationship that
> > allows
> > cwnd convergence with conventional CCAs, MDS aims to reduce
> > queueing
> > delay, and increase utilization for MDS CCAs, while supporting the
> > transition to scalable CCAs.
> > -----
> > 
> > I find the results interesting, and am looking to determine if this
> > idea is worth pursuing further.  Any feedback on the approach,
> > proposed
> > signaling mechanism or anything else is welcome.
> > 
> > Thanks!
> > Pete Heist
> > 
> > PS- I've also posted this to ICCRG in the IRTF, so apologies to
> > anyone
> > getting hit with it twice.
> > _______________________________________________
> > Ecn-sane mailing list
> > Ecn-sane@lists.bufferbloat.net
> > https://lists.bufferbloat.net/listinfo/ecn-sane
>