One of next year's probable projects is to incorporate the features in
minstrel-blues into mainline minstrel. Given just how hard it is to see
wins in this area I'd propose renaming the thing to something else so a
given box can be easily A/B switched and compared between regular
minstrel and whatever emerges from that effort.

BARD - Basic Adaptive Rate Decimator?

There are quite a few things in rate control worth looking at.

Also, the ath9k retries figure is still at a really high value. I long
ago cut it to < 8 (I think I used 6), but never pushed up the patch
(hard to measure) and is probably responsible for a few of the whiskers
of excess jitter here. A fixed number of retries is basically the wrong
thing, anyway... should end up time/airtime based vs the workload.

Still, testing a ubntlite today at 25 feet and 3 walls, it's nice to see
the ath9k outperform the ath10k in terms of bandwidth, be comparable in
terms of latency, and superior in terms of jitter.

OSX client ->

ath9k: HT20, channel 11
ath10k: VHT80, channel 153 (I think)