[ath9k-devel] Very large datagram loss

[Felix Fietkau <nbd@openwrt.org>]

On 2012-06-14 4:37 AM, Stephen Donecker wrote:
> Hi,
> 
> I am experiencing very high datagram loss between two identical cards
> communicating over a single spacial stream in adhoc mode. Using
> minstrel_ht rate control the tx bitrate settles at MCS7 150Mbps, and I
> get the following iperf results.
> 
> # iperf -c 192.168.11.2 -p 7777 -u -b 150m -t 10 -i 1
> ------------------------------------------------------------
> Client connecting to 192.168.11.2, UDP port 7777
> Sending 1470 byte datagrams
> UDP buffer size:  160 KByte (default)
> ------------------------------------------------------------
> [  3] local 192.168.11.1 port 43287 connected with 192.168.11.2 port 7777
> [ ID] Interval       Transfer     Bandwidth
> [  3]  0.0- 1.0 sec  16.1 MBytes   135 Mbits/sec
> [  3]  1.0- 2.0 sec  16.3 MBytes   137 Mbits/sec
> [  3]  2.0- 3.0 sec  16.3 MBytes   137 Mbits/sec
> [  3]  3.0- 4.0 sec  15.9 MBytes   133 Mbits/sec
> [  3]  4.0- 5.0 sec  16.4 MBytes   138 Mbits/sec
> [  3]  5.0- 6.0 sec  16.0 MBytes   134 Mbits/sec
> [  3]  6.0- 7.0 sec  16.2 MBytes   136 Mbits/sec
> [  3]  7.0- 8.0 sec  16.0 MBytes   135 Mbits/sec
> [  3]  8.0- 9.0 sec  16.2 MBytes   136 Mbits/sec
> [  3]  0.0-10.0 sec   161 MBytes   135 Mbits/sec
> [  3] Sent 115053 datagrams
> [  3] Server Report:
> [  3]  0.0-10.0 sec  92.0 MBytes  77.0 Mbits/sec   0.262 ms 49459/115052
> (43%)
> [  3]  0.0-10.0 sec  1 datagrams received out-of-order
> 
> When I set minstrel_ht to fixed_rate MCS7 I get the following iperf results.
> 
> # iperf -c 192.168.11.2 -p 7777 -u -b 150m -t 10 -i 1
> ------------------------------------------------------------
> Client connecting to 192.168.11.2, UDP port 7777
> Sending 1470 byte datagrams
> UDP buffer size:  160 KByte (default)
> ------------------------------------------------------------
> [  3] local 192.168.11.1 port 55776 connected with 192.168.11.2 port 7777
> [ ID] Interval       Transfer     Bandwidth
> [  3]  0.0- 1.0 sec  17.4 MBytes   146 Mbits/sec
> [  3]  1.0- 2.0 sec  17.4 MBytes   146 Mbits/sec
> [  3]  2.0- 3.0 sec  17.6 MBytes   147 Mbits/sec
> [  3]  3.0- 4.0 sec  17.4 MBytes   146 Mbits/sec
> [  3]  4.0- 5.0 sec  17.5 MBytes   147 Mbits/sec
> [  3]  5.0- 6.0 sec  17.6 MBytes   148 Mbits/sec
> [  3]  6.0- 7.0 sec  17.4 MBytes   146 Mbits/sec
> [  3]  7.0- 8.0 sec  17.4 MBytes   146 Mbits/sec
> [  3]  8.0- 9.0 sec  17.8 MBytes   149 Mbits/sec
> [  3]  0.0-10.0 sec   175 MBytes   147 Mbits/sec
> [  3] Sent 124930 datagrams
> [  3] Server Report:
> [  3]  0.0-10.2 sec  10.2 MBytes  8.39 Mbits/sec   4.527 ms
> 117609/124920 (94%)
> [  3]  0.0-10.2 sec  1 datagrams received out-of-order
> 
> In either case the overall throughput and datagram loss percentages are
> terrible. I believe overall the datagram loss should be less than a few
> percent.
Your expectations seem quite wrong. A PHY rate of 150 Mbit/s will not
get you anywhere near 150 Mbit/s UDP throughput, there's lots of 802.11
related overhead inbetween.
To check some stats on tx rate and retransmissions, take a look at
/sys/kernel/debug/ieee80211/phy0/netdev:wlan0/stations/*/rc_stats
(without fixed-rate).
One reason why the fixed-rate iperf looks so much worse is that it
disables aggregation.

- Felix