[Cake] Advantages to tightly tuning latency

[Dave Taht]

On Thu, Apr 23, 2020 at 4:05 PM Toke Høiland-Jørgensen <toke at redhat.com> wrote:
>
> Maxime Bizon <mbizon at freebox.fr> writes:
>
> > On Thursday 23 Apr 2020 à 20:35:15 (+0200), Toke Høiland-Jørgensen wrote:
> >
> >> I meant more details of your SOC platform. You already said it's
> >> ARM-based, so I guess the most important missing piece is which (Linux)
> >> driver does the Ethernet device(s) use?
> >
> > - Marvell Kirkwood, mv643xx_eth driver
> > - Marvell A8k, mvpp2 driver
>
> No native XDP support in any of those, unfortunately :(
>
> >> Yup, I think so. What does your current solution do with packets that
> >> are destined for the WiFi interface, BTW? Just punt them to the regular
> >> kernel path?
> >
> > yes, but that won't fly anymore for 11ax rates
>
> Indeed, that was partly why I asked :)
> Got any plans?

I would really love to get something going out of this initiative:

https://pointer.ngi.eu/pages/ngi-pointer-opencalls

... either with marvell wifi, 802.11ax, ath11k, mt76... gpon onts,
just something, anything fq-ing at the very least, over fiber of any
sort, would be a start.

is there anyone out there with some spare time and needs some eu dollars?

I tend to disagree that we need massive offloading on 802.11ax. We
need firmware on the chip that does per station scheduling, for the
rest of the queue management a smarter host cpu in the a72 class more
than suffices, especially if the packets are arriving gso'd.

>
> -Toke

>


--
Make Music, Not War

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-435-0729