[Make-wifi-fast] netnext, ubuntu fq_codel+airtime for ath9k, fq_codel for ath10k

Dave Taht dave.taht at gmail.com
Sun Oct 9 13:44:05 EDT 2016


I have updated toke's patch set to apply on top of net-next.

You can get pre-built kernel .debs and the patches from:

http://www.taht.net/~d/airtime-8/

These were built for ubuntu 16.04.

I've also supplied a sample hostapd.conf file for x86. I found it
necessary in some cases to "kick" the ath9k card into 5ghz mode by
manually running iwconfig essid whatever channel 36 before hostapd
would start. (I did not get HT40 to work on this card, but I never had
before)

Tested on the pcengines apu2d2 (ath10k), and a Qualcomm Atheros AR9462
in a laptop acting as AP... with up to 5 clients, mostly from the last
working build for lede.

Note, these are "unofficial" - my main purpose was to test crypto and
make sure that the stuff didn't crash.

1) I did not preserve authorship in these (how do you apply git
patches with fuzz?)
2) I moved back the fq_codel defaults to the same as mainline, rather than
the relaxed defaults others have been using. For all I know this will mess
with codel, but there are a few other approaches to try there.
3) enabled the fq_codel stuff for the ath10k cards I have
4) I did not apply toke's patch to make codel gentler at lower rates

As these are based on net-next, I wouldn't risk a machine you care
about to this! I have got an occasional oops from my graphics card in
the laptop.

* Tested on the ath10k and ath9k for several days
**   not having airtime fairness on the ath10k hurts, but for single station
     benchmarks, it's quite lovely.
** ath9k -> elsewhere is nice at both 2.4ghz and 5ghz.
   http://www.taht.net/~d/airtime-8/upisgood.png
   maybe it's too good, have to look at codel harder
** OSX appears to be cheating somehow
** stats seem to be wrong in this case?

root at nemesis:/sys/kernel/debug/ieee80211/phy0/netdev:wlp3s0# cat aqm
ac backlog-bytes backlog-packets new-flows drops marks overlimit
collisions tx-bytes tx-packets
2 0 0 747 0 0 0 3 235451 748

but ok in this case:

root at nemesis:/sys/kernel/debug/ieee80211/phy0/netdev:wlp3s0/stations/48:d7:05:c0:9a:d9#
cat aqm
tid ac backlog-bytes backlog-packets new-flows drops marks overlimit
collisions tx-bytes tx-packets
0 2 0 0 245103 470812 0 0 2 1232820484 1844136




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


More information about the Make-wifi-fast mailing list