[Cerowrt-devel] [Make-wifi-fast] wifi airtime fairness patches could use eyeballs and testing

[Dave Taht]

Toke and I have now done a great deal of testing and exploration of
the issue with leaving "fq" enabled on the ath9k (running with and
without felix's patch).

The only thing we've found that "moves the bar" is the presence or
absence of crypto. With crypto + fq on, on the wndr 3800, we see a
significant performance variation of 32-80mbit as the number of flows
goes from 12 to 1.

without crypto, with or with fq enabled, we are able to reliably get
120Mbits on our test setup.

The patterns of the degradation are also often variable, we will have
long periods where things are quite stable, and other 10-20 sec
periods where throughput drops or stops almost entirely.

It would be nice to get confirmation that the fq is working without
crypto enabled.

Ideas as to why crypto chops off so much throughput in the first place
(from 120mbit to 80), and as to what else to poke into with crypto on,
would be nice.