FWIW, I tried this on the 2.2 release and it did not disable WED. Not sure why and I did not spend a lot of time chasing it down. WED was disabled after I added options mt7915e wed_enable=N to /etc/modules.conf as I have been recommending.
I'm not able to load the 2.4 release right now, so if someone can confirm that removing wed_enable=1 from /etc/modules.d/mt7915e does in fact disable WED, that would be great
Fair question Yes, I reboot any time I toggle the WED setting change. I can confirm that when WED is disabled, I no longer see the WO (only WA/WM) firmware getting loaded during reboot:
Conversely, when I remove the wed_enable=0 argument and leave only mt7915e in /etc/modules.d/mt7915e, you'll see that WED becomes enabled again (confirmed by the presence of WO firmware loaded during boot):
My hunch is that the weirdness with WED not truly being disabled even when removing wed_enable=1 is causing testers to believe they are actually testing with the new AQL settings, but in fact are not because of WED still being enabled.
Once testers truly disable WED and start testing against AQL, I think the latency picture will look much brighter regardless of the end user's device type
Up next... I am going to re-roll your r2.4 build with another mac80211 patch to modify the codel target and interval to some more sane values I've been using for a long time now. I'll do another WED disabled benchmark and post the result here...
So, this time I really disabled WED by also modifying /etc/modules.conf to include options mt7915e wed_enable=N
Got fairly high CPU usage but great bufferbloat results (A), all without SQM which in my opinion is not really needed unless we're amining at a perfect A+.
OnePlus:
PC latency is always a touch lower, this time just 10ms for up and down.
I think I can help because I'm getting consistent results and I have a good range of devices I can use for testing.
I also took the chance to test with the temporary AQL settings from this message. Just those, no SQM configured. I don't see a relevant change, but I believe I actually need to enable SQM, can you please confirm? I always do 2/3 runs and try to average the numbers.
The only weird change is on the OnePlus: I get much lower latency with download, but the same as before for upload. Only happens on this device so I don't know.
Please let me know if I actually have to enable SQM, otherwise I think I'm done testing for today
Even if CPU load is comparable, WiFi performance is destroyed. Unless I did something stupid (which is entirely possible) I am not going to consider SQM+WED off.
Latency is better but the speed cost is not worth it in my opinion.
Regarding SQM, if you set inteface on wifi, upload and download bandwidth should be reversed...
Upload from access point to device.
Download, from your device to access point