Re: [Cake] [Make-wifi-fast] Airtime fairness patch merged upstream, and the Turris Omnia debloated

[Dave Taht <dave.taht@gmail.com>]

On Fri, Dec 23, 2016 at 5:32 PM, Aaron Wood <woody77@gmail.com> wrote:
> Toke,
>
> Do you know if it's made it into the LEDE nightlies, yet?  I loaded up the
> Dec 22 nightly onto my WNDR3800.  Running rrul I'm getting about 10-30ms of
> latency over wired on the same link (and wifi is definitely the limit, as

I'm assuming you meant wireless above?

> it's using piece_of_cake at 135M down, 12M up (and I'm seeing 0 measurable

I really don't recommend cake at much greater than 60Mbits on this
hardware inbound.

> additional latency from piece of cake).

rrul stresses out both sides of the wifi connection AND all the Qos queues.

rrul_be stresses out both sides but only one queue.

rrul remains generally awful on wifi unless your dscp markings have
been eliminated in transit.

for rrul_be - unless you've got the fq_codel code on both sides, you
aren't going to see as much benefit, and even if you are, the double
buffering on both sides leads to a typical latency of about 10ms each
way in this scenario. (so if you got it, you are inline with our
results).

I generally switched to using tcp_nup and tcp_ndown (with varying
numbers of flows), as well as the various rtt_fair_var tests against
multiple stations, to evaluate the wifi changes.

It's not clear to me if the quantum and codel targets were restored to
their defaults in this release (300->1514, 20ms -> 5ms)



> -Aaron
>
>
>
> On Mon, Dec 19, 2016 at 2:31 PM, Toke Høiland-Jørgensen <toke@toke.dk>
> wrote:
>>
>> The airtime fairness patch for ath9k has been accepted into the mainline
>> Linux kernel, and is queued to be merged for 4.11 (i.e. the next merge
>> window after the current 4.10 cycle completes).
>>
>> The patch has also been accepted into LEDE and is currently in Felix'
>> staging tree at
>> https://git.lede-project.org/?p=lede/nbd/staging.git;a=summary - so if
>> nothing surprising shows up, it should make it into the regular LEDE
>> nightlies before too long :)
>>
>>
>> In related news, the Turris Omnia had the WiFi queue restructure patches
>> added in the latest update; and the ath10k chip it ships with seems to
>> have the required hardware support, so both WiFi devices in the Omnia
>> now have nicely debloated queues. They also switched to sqm-scripts
>> (from wondershaper that they shipped before), and added Cake :)
>>
>> -Toke
>> _______________________________________________
>> Make-wifi-fast mailing list
>> Make-wifi-fast@lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>
>
>
> _______________________________________________
> Cake mailing list
> Cake@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/cake
>



-- 
Dave Täht
Let's go make home routers and wifi faster! With better software!
http://blog.cerowrt.org