Hi everyone. I'm running also an APU2 board with an Atheros wlan chipset:
athn0 at pci5 dev 0 function 0 "Atheros AR9281" rev 0x01: apic 5 int 16 athn0: AR9280 rev 2 (2T2R), ROM rev 22, address XX:XX:XX:XX:XX:XX I also used an ALIX board with an Atheros wlan chipset (AR9280). >From OpenBSD 6.0 to 6.6 (-stable) (iirc OpenBSD 5.9 had a slightly better performance) the wlan performance didn't change for me. I use the following test setup: PC <--- 100mbps / 1gbps lan ---> ALIX / APU2 board <--- Atheros wlan ---> Notebook # cat /etc/hostname.athn0 media autoselect mode 11g mediaopt hostap chan 3 nwid test wpa wpakey wpasecret wpaprotos wpa2 wpaciphers ccmp wpagroupcipher ccmp powersave up For simplicity I disable pf and use a bridge interface that uses the lan and wlan interfaces. Also I always send files with scp from my PC to my notebook for performance measurements. The ALIX board has a wlan throughput of about 2.4 - 3.2 mbps while the APU2 has a wlan throughput of about 12 - 14 mbps. I noticed by running top -S, that if I transfer lets say a 1000MB file, then for both boards softnet has a CPU usage of 100%, thus I suspected that the wlan bandwidth for an Atheros chip correlates to the CPU usage of softnet (because the APU2 board has a better CPU it has a higher wlan throughput). If I send a 1000MB file from one lan interface to another lan interface, then softnet has a CPU usage of about 20% and the file transfer has a throughput of almost 1gbps. Thus I suspect the high CPU usage of softnet has something to do with the TX performance of the atheros driver implementation. For a quick test I enabled pf and used traffic sharping to reduce the wlan troughput: queue std on "athn0" bandwidth 5M max 10M default Now when I transfer a file softnet has a CPU usage of about 70-73% for the APU2 board (the wlan throughput is about 10 - 11 mbps). Also I found an interesting performance report, that pfSense could reach about 90 mbps with the same hardware that I have (APU2 board with the same wlan card) see [1], unfortunately I don't have the time to verify it. If that's true, then the next step I want to do is a diff of the driver implementations and hopefully understand why pfSense has a much higher throughput. [1] https://teklager.se/en/knowledge-base/compex-wle200nx-wle600vx-benchmark/ best regards, Mat -- Sent from: http://openbsd-archive.7691.n7.nabble.com/openbsd-user-misc-f3.html
