[LEDE-DEV] [PATCH] ath10k: Re-enable intermediate softqueues for all devices

[Dave Taht]

Baptiste Jonglez <baptiste at bitsofnetworks.org> writes:

> On 19-09-17, Toke Høiland-Jørgensen wrote:
>> Baptiste Jonglez <baptiste at bitsofnetworks.org> writes:
>> 
>> > Hi,
>> >
>> > On 14-09-17, Toke Høiland-Jørgensen wrote:
>> >> Toke Høiland-Jørgensen <toke at toke.dk> writes:
>> >> 
>> >> > The upstream ath10k driver disables the intermediate softqueues for some
>> >> > devices. This patch reverts that behaviour and always enables the
>> >> > softqueues (and associated bufferbloat fixes). We have had reports of people
>> >> > running this with good results:
>> >> > https://lists.bufferbloat.net/pipermail/make-wifi-fast/2017-September/001497.html
>> >> 
>> >> So there seems to be some issues with QCA9980 with this patch. See
>> >> https://lists.bufferbloat.net/pipermail/make-wifi-fast/2017-September/001505.html
>> >
>> > I have updated my Archer C2600 to 53839da46e a few days ago (your patch
>> > was just merged) and did not notice any particular performance issue.  The
>> > Archer C2600 also has a QCA9980.
>> 
>> Cool, thanks for testing! The other performance problems that were
>> reported also seem to be fixed, so no need to revert the patch after all :)
>
> OK, good news :)
>
> I did stress the Archer C2600 but forgot to report about it, everything
> looked normal (up to ~450 Mbps on 5 GHz).

Well, the core portion of the patch's intent was to reduce
bufferbloat-related latency at low rates, without costing throughput.

I am curious (if artificially limited somehow), if any results even
vaguely similar to to the initial effort and final paper - are coming
out of this version against this chipset.

initial POC: http://blog.cerowrt.org/post/fq_codel_on_ath10k/

final paper: https://www.usenix.org/system/files/conference/atc17/atc17-hoiland-jorgensen.pdf

> _______________________________________________
> Lede-dev mailing list
> Lede-dev at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/lede-dev