On Tue, Mar 30, 2021 at 11:06:43PM -0400, Scott Bennett wrote: > However, my laptop with AR9287 was noticeably worse with this diff (dropped > pings, stuttering keystrokes in interactive ssh session, estimated 20 > minutes to scp(1) a 20M file...). The combination of apu2 with diff and my > laptop sans diff is giving me good results though :) > > athn0 at pci2 dev 0 function 0 "Atheros AR9287" rev 0x01: apic 2 int 17 > athn0: AR9287 rev 2 (2T2R), ROM rev 4, address 74:de:2b:xx:xx:xx
This device (AR9287) has known issues in 11n mode already: https://marc.info/?t=158992287500008&r=1&w=2 Granted, that is a sample size of one, and unlike the person who filed that bug report you didn't have problems before the RA patch. But perhaps these issues are somehow related? I don't have this hardware, and I cannot tell what's wrong with it. My best guess is that this device doesn't report retries in the same way as the 9280 does, so RA ends up picking a rate that's too high. But that's just a wild guess. Can you please try the following: Set a fixed Tx rate and figure out the loss/throughput characteristics of each, while the laptop stays put at the same place relative to the AP. The supported Tx rates for this device are grouped into two groups: Group1 : MCS 0 (low) to MCS 7 (high) Group2 : MCS 8 (low) to MCS 15 (high) Start with a fixed Tx rate of MCS 0: ifconfig athn0 media HT-MCS0 mode 11n Now do you see dropped pings, stuttering keystrokes, slow transfers? Ideally you'd show ping packet loss at the default size and at ping -s 1500, plus the transfer rate when sending a big file from the laptop towards the AP. Note that scp isn't a good tool for measuring Tx performance; rsync or a benchmarking tool such as tcpbench or iperf are suited better for this. Then go up and repeat the test at each rate: ifconfig athn0 media HT-MCS1 mode 11n ... ifconfig athn0 media HT-MCS7 mode 11n Once you've reached MCS 7, move on the next group: ifconfig athn0 media HT-MCS8 mode 11n ... ifconfig athn0 media HT-MCS15 mode 11n It's a bit tedious, but if you could give me a test result for all the 16 supported MCS we might be able to find a fix. Thanks!
