[Solved] 802.11ax worse than 802.11ac with mt76 driver?

Cannot confirm 802.11ax being worse than 802.11ac with mt76 driver. In my tests behind 2 walls, ax is consistently faster than ac

Device: D-link-dap-x1860 Not yet officially supported, but a patch has been sent to the mailing list.

Stock OEM software:

  • (A) comparison between Stock and OpenWRT
  • (B) detailed Iperf3 bidirectional test behind 2 walls

RolandMagico's OpenWRT Snapshot from 2022-12-22:

  • (A) comparison between Stock OEM and OpenWRT
  • (B) detailed Iperf3 bidirectional tests with low and high number of client streams (behind 2 walls)

Here some excerpts:

The best news first:

  1. OpenWRT has higher range than stock firmware. Whereas with stock firmware I had a location in my house (behind two walls and far distance), where my phone constantly lost connection to the repeater (I assume it was because stock firmware automatically tries to switch between 2.4G and 5G, when low signal strength is detected), now with OpenWRT, there are no connection issues at all when I am with my phone in this specific location. I simply configured two separate wifi SSIDs (one for 2.4G(ax) and one for 5G(ax) and my phone only connects to one of them (roaming disabled). It is rock stable. I was able to switch from 144p to 720p youtube video quality :smiley:

  2. Low amount of laggs. I was playing computer games (Heroes of the Storm from Blizzard) yesterday for a few hours and had zero laggs. (At this specific location, I rarely had laggs with Stock firmware either, so actually no change here). Client is located behind two walls. It is the location from my iperf3 measurements down below.

The bad news:

  • Speeds in certain distance ranges are lower than stock OEM (but for me that is ok, since I gain control over configuration, (hopefully) more stability, increased range and the potential for future improvements in OpenWRT)

Here a glimpse of my Iperf3 measurements:

Remarks about iperf3 measurements:

  • Higher number of client streams seems to have low impact on overall (mean) performance.
  • with 8 streams, OpenWRT performance with bidirectional tests is roughly comparable to stock OEM, but OpenWRT has LESS retries (In one of my tests, OpenWRT had 20-50, whereas Stock OEM had 800-1300). Stock OEM performs a lot better in simple tcp tests though.
  • Many client streams lead to whacky performance (Similar on both stock OEM and OpenWRT), as a higher number of clients has a very strong impact on median speed and standard deviation. Speed per stream varies wildly. Some streams can capture full speed, whereas the speed of other clients drops rapidly in comparison. There seems to be no bandwidth fairness or alternative mechanism that would keep outliers in check.