While testing a memory leak patch from current on up-to-date netbsd-9, I found strange behavior with a urtwn(4). On a 2006 MacBook Pro which otherwise works well, I have been using a urtwn(4) interface. If I run iperf3 with streams in both directions and with v4 and v6, I get about 20 Mb/s for three cases and about 500 kb/s for transmitting v4. From using a homemade blaster to discard, tcpdumping, and visualizing the TCP behavior with xplot, it looks like packets are lost. There's a lot of SACK and packets are retransmitted. Sometimes that's fast, and sometimes there are timeouts. I see a lot of dupacks in the netstat -s. I'm guessing that somehow some transmitted packets just aren't sent. My wifi network is otherwise working fine; other hosts to/from all have expected behavior. I'm trying this on a RPI3 with netbsd-9 from October, with a different urtwn(4) device, and not seeing the issue. So if you have a urtwn(4) device, and are in the mood to test with iperf, I am curious if "iperf3 -c foo -4" is slow, while -6, or either with -R is reasonable. (Run "iperf3 -s" on foo.)
Attachment:
signature.asc
Description: PGP signature