Re: [tsvwg] feedback and thoughts L4S / SCE

[Jonathan Morton <[e-mail]>]

> On 28 Nov, 2020, at 4:20 pm, Mikael Abrahamsson <swmike=[e-mail]> wrote:
> 
>> As it turns out there is a nssfq_codel modul for qualcomm`s nss network acellerator cores:
>> see https://forum.openwrt.org/t/ipq806x-nss-drivers/12613/2160
>> 
>> Is that a sufficient response to put tge hypothesis to rest that acelleration techniques do not offer fq_codel?
> 
> This is great news. Do you have more examples?
> 
> I looked over the devices list with that SoC and most of them were 130EUR here (or more). Pretty high-end devices.

The SoC involved here is 5 years old.  Think about that for a moment.

Several of the devices on that list are no longer on the market, because they've been superseded.  So this is hardware that was moderately high-end (in a consumer CPE context) several years ago.  The only reason we're hearing about this nssfq_codel functionality right now is because the method of turning it on finally got reverse-engineered in the open-source world, no thanks to the hardware vendor.  I wish they would stop shooting themselves in the foot on that score.

I have no doubt that similar functionality is actually commonplace in the better CPE SoCs now, certainly in later Qualcomm products.  It's just not emphasised in the spec sheets because the marketing department doesn't understand it, and getting access to it from open source is like getting blood out of a stone.  But it is *there*.

Sebastian pointed out that this constitutes an existence proof for hardware-accelerated fq_codel functionality in reasonably-priced network-edge devices.  We also know that Approximate Fairness technology is available in ISP-grade hardware.  CPE vendors are actively promoting Airtime Fairness technology on the wifi side, which I happen to know comes along with fq_codel as a side effect.

At this point, any claim that fq_codel is not widely deployed in relevant contexts is laughable.

 - Jonathan Morton