[Make-wifi-fast] trouble on tid 5

[Dave Taht]

I had been testing ath9k a bit with fq_codel a bit with patch tim
supplied me, and I noticed an issue with the VI queue. Might be across
more queues than that, but definately VI.

examples see around packet number 4394, 8638, and like 139144 - in the cap:

http://www.taht.net/~d/carnage.cap.gz

the essential dialog:

station: Here you go, frames 60 through 68. (IP x.y.19.15)

AP: (IP x.y.18.1) I didn't get frames 78-123

station: Here you go, frames 60 through 68, at a different rate:

AP no, you idiot! I didn't get frames 78-123

The cycle repeats XX times for a total of YY microsecs, before the
driver gives up...

I did not poke much further, thought I'd supply an aircap so folk
could look at what one looked like.  Me, all I wanted to do was fiddle
with the Ack Policy field but haven't got to it yet.

...

data generated via aircapping a rrul test between tim's fq_codel patch
and a cerowrt box. The box seemingly misunderstanding the AP is tim's
side....

...


The worst part of this is that' it all for acks. All we had to do to
keep the tcp flow going was 1 ack - frame XX, and tcp would have
handled it ok...

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