Theoretical (advertised) speeds are 400Mbps (2.4GHz) and 867Mbps (5GHz)
Tested using iperf3 client on a phone (wireless to GL B-2200) and iperf3 server on a PC (wired to GL B-2200), trying to keep all things equal / settings out of box.
I can live even with OpenWrt speeds, but really curious what is behind such degradation of the speed and if anything can be done to restore parity.
2.4GHz radio throughput rose to 68/91 Mbps (only "Software flow offloading" checked), almost to the stock throughput
5GHz radio throughput rose to 362/415 Mbps (only "Software flow offloading" checked), but still lagging the stock throughput
Additionally enabling "Hardware flow offloading" had little impact, the results were very close to the ones with only "Software flow offloading" enabled.
I am afraid, I did not manage to [easily] locate this setting.
Ultimately, managed to test different combinations of "Packet steering", "Software flow offloading" and irqbalance enabled individually, as well their combinations.
2.4GHz download: "Packet steering" produces the highest speed (74Mbps), which is at par with stock firmware download speed (78Mbps). "Software flow offloading" is second best (68Mbps), almost all other options are below 50Mbps. 2.4GHz upload: all settings produce more than 90Mbps speed, e.g. irqbalance - 95Mbps, "Packet steering" - 93Mbps, "Software flow offloading" - 91Mbps". The highest speed is for the combo of irqbalance and "Packet steering" - 98Mbps.
Based on these results from 2.4GHz radio, I am leaning to keep "Packet steering" enabled.
5GHz download: "Packet steering" produces the highest speed (438Mbps), still below the stock firmware speed (605Mbps). The combo of irqbalance and "Packet steering" is second best (423Mbps). 5GHz upload: somewhat mixed bag. irqbalance + "Software flow offloading" + "Packet steering" win with 534Mbps speed. "Packet steeering" + "Software flow offloading" second best with 485Mbps. irqbalance alone produces 477Mpbs, and "Packet steering" alone - 467Mbps. Stock firmware is at 464Mbps.
For 5GHz, "Packet steering" again seems to be the optimal option (best for download, but only OK for upload).
All-in-all, having "Packet steering" enabled seems to raise OpenWrt Wi-Fi performance up to the level of firmware Wi-Fi performance for both radios, with 5GHz download being the only exception, lagging by some 166Mpbs the firmware download speed.
@pythonic Thanks a lot for pointing me to these settings!
I checked with cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor, the output was performance.
Interesting results at 5GHz - even with just packet steering you're equalling the stock upload; it's just download where you're losing a bit. BTW, the two 5GHz radios operate with different channels (wifi1 is the SoC 5GHz radio in low band, wifi2 is the qca9888 radio in high band) with different drivers and firmware so which radio were you testing against? While I wouldn't expect there to be any significant performance difference between OpenWrt's default ath10k-ct and the OEM QSDK ath10k drivers/firmware you wouldn't know without testing.
The only other thing that occurs to me is that the transition to DSA at 23.05 may be affecting performance a little (on the LAN switch side) perhaps affecting the 5GHz upload; you'd have to repeat the tests with 22.03.6 (which is still on swconfig) to confirm. Also worthwhile monitoring with htop to see whether 1 CPU is hitting limits
irqbalance can be quite good but sometimes you can achieve overall more consistent results by manually analysing and pinning irqs, though at the cost of effort.
My bad - I hadn't noticed the governor change had been backported to 23.05.
Yes, this is fair question, missed mentioning earlier. I was testing against the SoC IPQ4019 radio in low band (channels 36-64), not the QCA9888 (channels 104-165) because SoC radio was also the default firmware's 5GHz radio. Actually, did not bother to check how well the QCA9888 does with the OEM firmware, assuming this is only used for mesh (which I was not interested in). But probably can still be used for e.g. hosting one of the SSIDs?
You got me curious But it will take time to do that, I am afraid.
Could be due to DSA in 23.05.2 or maybe it is in general the price for upgrading to more recent Linux kernels, dunno. The stock firmware is based on Chaos Calmer.
Neither soft offloading, nor irqbalance, nor packet steering improved the download speeds. But packet steering improved 5.0GHz upload to 479Mbps. Each packet steering and irqbalance improved 2.4Ghz upload to 95-96Mbps.
Probably would stick with 22.03.6, at least while it is supported until April 2024. The main purpose for the device is to be dumb access point, hopefully not much damage from staying on the older OpenWrt version. Edit: 23.05.2 though gives higher speed with packet steering enabled, faster option.
