[Ecn-sane] sce ramp vs threshold

[Ecn-sane] sce ramp vs threshold

[Dave Taht]

My original understanding of sce was that a threshold function
sufficed for fq, and a ramp was needed for
single queues. Is that still the case?

I *reallly* want to see the cpu overhead of this stuff on cheap
hardware, so I'm fiddling with this:

https://github.com/dtaht/fq_codel_fast/commit/9e6ab76c81fd26f22806f19476481a1b92c9703e

-- 

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-205-9740

Re: [Ecn-sane] sce ramp vs threshold

[Jonathan Morton]

> On 8 Aug, 2019, at 7:17 pm, Dave Taht <dave.taht@gmail.com> wrote:
> 
> My original understanding of sce was that a threshold function sufficed for fq, and a ramp was needed for single queues. Is that still the case?

It's more the case - as it always has been - that a threshold function suffices as the minimum functional implementation, but a ramp function performs better even with FQ.  With a ramp, it is possible for the cwnd to stabilise *at* a balancing value rather than oscillating around it.

We're still exploring the solution space for single queues.  A ramp function might not be the best option there, but the kinked ramp shown in the slide deck seems to work okay when correctly tuned.

 - Jonathan Morton