Hello,
TLDR: 5.10 seems to have performance regressions, especially in upload (ax200 sending to AP)
direction with TCP.
Here's some results from testing ax200 on 5.4 and 5.10 kernels.
Kernels can be found here:
https://github.com/greearb?tab=repositories
I have a system with 4 ax200 NICs plus some ath10k radios (ath10k idle in this
test case). AP is ath11k 8x8 hawkeye.
5.4 (using out-of-tree backports ax200 driver since in-kernel driver was very unstable for us)
I am not sure the exact version I backported, it was back in 6/8/2020 of this year.
single station download: 900Mbps
4 stations download: 900Mbps
single station upload: 775Mbps
4 stations upload: 700Mbps
5.10 (we first tested in-kernel driver, and that showed poor upload, so then I tried
hacking the backports release/core56 ax200 driver into 5.10. Detailed testing results
below are from the hacked in backports driver)
single station download: 500Mbps
4 stations download: 900Mbps
single station upload: 450Mbps
4 stations upload: 400Mbps
UDP ran similarly. TCP with 350 streams ran similarly.
As always with wifi, there are a million things that could be weird
about my test case. If someone can do similar testing and report
results, please do so we can compare.
Thanks,
Ben
--
Ben Greear <[email protected]>
Candela Technologies Inc http://www.candelatech.com