Re: rt61pci - bad performance

From: Andreas Hartmann
Date: Mon Aug 16 2010 - 13:02:20 EST

Helmut Schaa wrote:
Added rt2x00 mailinglist to CC ...

Am Saturday 14 August 2010 schrieb Andreas:
Helmut Schaa wrote:
Hi Andreas,

Am Freitag 13 August 2010 schrieb Andrew Morton:
(cc's added)

On Sun, 08 Aug 2010 11:49:49 +0200
Andreas<andihartmann@xxxxxxxxxxxxxxx> wrote:

[...]

wlan0 IEEE 802.11bg ESSID:"--------"
Mode:Managed Frequency:2.412 GHz Access Point: some AP
Bit Rate=1 Mb/s Tx-Power=5 dBm
Retry long limit:7 RTS thr:off Fragment thr:off
Encryption key:off
Power Management:off
Link Quality=38/70 Signal level=-72 dBm
Rx invalid nwid:0 Rx invalid crypt:0 Rx invalid frag:0
Tx excessive retries:0 Invalid misc:0 Missed beacon:0

The throughput is measured with ping -f -s 7000 and xosview -n.

This doesn't look like an appropriate way to measure the throughput. You
should use something like iperf [1] or netperf [2] for your measurements
to get more accurate results.

If I'm using ndiswrapper with the windows driver, first of all, I can
see additional information in iwconfig:

wlan0 IEEE 802.11g ESSID:"--------"
Mode:Managed Frequency:2.412 GHz Access Point: some AP
Bit Rate=54 Mb/s Tx-Power:20 dBm Sensitivity=-121 dBm
RTS thr=2347 B Fragment thr=2346 B
Encryption key:some key Security mode:restricted
Power Management:off
Link Quality:62/100 Signal level:-56 dBm Noise level:-96 dBm
Rx invalid nwid:0 Rx invalid crypt:0 Rx invalid frag:0
Tx excessive retries:0 Invalid misc:0 Missed beacon:0


There is a switch for sensitivity (which is not supported with rt61pci)
and the link quality compared with ndiswrapper is worse (38% to 62%).

I wouldn't trust the link quality values that much, the calculation in rt61pi
is most likely different from what the windows driver does. So it is not
really comparable.

I detected the problem using tunneled ssh-x-sessions and during copying
of data. I'm not really interested in the link-quality - I just need a
high performance :-).

The following is remarkably too:
ndiswrapper uses a Tx-Power of 20 dBm, rt61pci only 5 dBm. I don't know,
why rt61pci uses 5 dBm. It's a hard limit and I can't set it on a value
higher than 5 unless the driver is patched. Nevertheless, setting a
higher value (of 20 dBm) by patch does not mean to get a better performance.

Could you elaborate please? Did you actually try to patch it or is this just
an assumption?

see my other mail!

Ndiswrapper shows an encryption key, rt61pci not. Does it mean, that
rt61pci doesn't use hardware encryption?

hw crypto should be enabled by default in rt61pci, however, I don't know
if it is actually working ;)

How can I see if it's working?

You can add a printk to rt61pci_fill_rxdone, something like:

diff --git a/drivers/net/wireless/rt2x00/rt61pci.c b/drivers/net/wireless/rt2x00/rt61pci.c
index e539c6c..aa1aafd 100644
--- a/drivers/net/wireless/rt2x00/rt61pci.c
+++ b/drivers/net/wireless/rt2x00/rt61pci.c
@@ -2023,6 +2023,7 @@ static void rt61pci_fill_rxdone(struct queue_entry *entry,
rxdesc->flags |= RX_FLAG_DECRYPTED;
else if (rxdesc->cipher_status == RX_CRYPTO_FAIL_MIC)
rxdesc->flags |= RX_FLAG_MMIC_ERROR;
+ printk(KERN_INFO "rt61pci_fill_rxdone: %x\n", rxdesc->cipher_status);
}

/*


With ndiswrapper, the rt61pci-chip achieves a throughput of 2,6 MBytes/s
- that's about 1 MByte/s more than rt61pci.

I have to say, that the difference between rt61pci and ndiswrapper gets
worse if the link quality is getting more badly. Or in other words:
ndiswrapper handles bad connections better then rt61pci.


Do you have any idea to get rt61pci working as fast as ndiswrapper?

Please run proper measurements first and post the results again.

I did some measurements with netperf (TCP_STREAM):


ndiswrapper
===========

(OpenSuSE 11.2 2.6.31.13-21):
download
average min max
20,88 19,02 22,19 MBit/s (6 runs)

upstream
average min max
21,46 18,84 22,26 MBits/s (7 runs)


OpenSuSE 11.3 (2.6.34-12-desktop)
download
average min max
21,41 20,51 22,51 MBit/s (16 runs)


upstream
average min max
error


rt61pci (patched - compat-wireless-2010-07-20)
==============================================

OpenSuSE 11.3 (2.6.34-12-desktop)
download
average min max
15,54 12,4 17,19 MBit/s (25 runs)

upstream
average min max
13,54 12,1 14,04 MBits/s (7 runs)

Hmm, ok that's quite a difference. Could you please mount debugfs
(mount -t debugfs none /mnt), rerun the test and attach the contents
of /mnt/ieee80211/phy0/stations/XX\:XX\:XX\:XX\:XX\:XX/rc_stats
afterwards (XX:XX:XX:XX:XX:XX is the BSSID your connected to).

Wel, I did some tests. I tried to get same conditions (what is not as easy). I will show here some results, which seem to be typical to me.


downstream
==========

netperf -t TCP_SENDFILE -H client
TCP SENDFILE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to client port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec

87380 16384 16384 10.36 16.38


/sys/kernel/debug/ieee80211/phy0/stations/xx:xx:xx:xx:xx:xx # cat rc_stats
rate throughput ewma prob this prob this succ/attempt success attempts
1 0.7 76.2 100.0 0( 0) 11 11
2 0.0 0.0 0.0 0( 0) 0 0
5.5 0.0 0.0 0.0 0( 0) 0 0
11 0.0 0.0 0.0 0( 0) 0 2
6 0.0 0.0 0.0 0( 0) 0 32
9 0.0 0.0 0.0 0( 0) 551 1574
12 0.0 0.0 0.0 0( 0) 2096 6862
18 11.3 69.9 66.6 0( 0) 18047 25158
t 24 13.4 62.9 100.0 0( 0) 29100 42883
T P 36 28.2 92.8 100.0 1( 1) 135030 175797
48 4.4 11.3 0.0 0( 0) 361 3646
54 0.8 1.8 0.0 0( 0) 55 1727

Total packet count:: ideal 8917 lookaround 991



netperf -t TCP_STREAM -H client
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to client port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec

87380 16384 16384 10.39 16.57


/sys/kernel/debug/ieee80211/phy0/stations/xx:xx:xx:xx:xx:xx # cat rc_stats
rate throughput ewma prob this prob this succ/attempt success attempts
1 0.7 76.2 100.0 0( 0) 11 11
2 0.0 0.0 0.0 0( 0) 0 0
5.5 0.0 0.0 0.0 0( 0) 0 0
11 0.0 0.0 0.0 0( 0) 0 2
6 0.0 0.0 0.0 0( 0) 0 32
9 0.0 0.0 0.0 0( 0) 551 1614
12 0.0 0.0 0.0 0( 0) 2096 7047
t 18 12.8 79.2 80.0 0( 0) 18647 25949
24 9.9 46.7 100.0 0( 0) 29439 44023
T P 36 29.0 95.6 100.0 1( 1) 141588 183495
48 5.2 13.3 50.0 0( 0) 380 3781
54 12.4 28.6 100.0 0( 0) 60 1797

Total packet count:: ideal 6867 lookaround 763




upstream:
========

netperf -t TCP_MAERTS -H client
TCP MAERTS TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET client port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec

87380 16384 16384 10.56 13.19


/sys/kernel/debug/ieee80211/phy0/stations/xx:xx:xx:xx:xx:xx # cat rc_stats
rate throughput ewma prob this prob this succ/attempt success attempts
1 0.7 76.2 100.0 0( 0) 11 11
2 0.0 0.0 0.0 0( 0) 0 0
5.5 0.0 0.0 0.0 0( 0) 0 0
11 0.0 0.0 0.0 0( 0) 0 2
6 0.0 0.0 0.0 0( 0) 0 32
9 0.0 0.0 0.0 0( 0) 551 1723
12 6.8 61.8 100.0 0( 0) 2122 7635
P 18 16.0 98.6 100.0 0( 0) 21199 29108
t 24 16.6 78.2 100.0 0( 0) 44090 61942
T 36 29.1 95.7 100.0 1( 1) 183435 238929
48 0.0 0.0 0.0 0( 0) 446 4861
54 0.0 0.0 0.0 0( 0) 67 2340

Total packet count:: ideal 6696 lookaround 743



netperf -t TCP_MAERTS -H client
TCP MAERTS TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to client port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec

87380 16384 16384 10.39 13.38


/sys/kernel/debug/ieee80211/phy0/stations/xx:xx:xx:xx:xx:xx # cat rc_stats
rate throughput ewma prob this prob this succ/attempt success attempts
1 0.7 76.2 100.0 0( 0) 11 11
2 0.0 0.0 0.0 0( 0) 0 0
5.5 0.0 0.0 0.0 0( 0) 0 0
11 0.0 0.0 0.0 0( 0) 0 2
6 0.0 0.0 0.0 0( 0) 0 32
9 0.0 0.0 0.0 0( 0) 551 1534
12 0.0 0.0 0.0 0( 0) 2096 6669
18 15.5 95.4 100.0 0( 0) 17433 24342
tP 24 20.9 98.3 100.0 0( 0) 28782 41742
T 36 28.9 95.2 100.0 1( 1) 128587 168213
48 0.0 0.0 0.0 0( 0) 341 3510
54 0.0 0.0 0.0 0( 0) 54 1658

Total packet count:: ideal 2073 lookaround 230


It's remarkably, that the upstream is 3 Mbit/s lower than the downstream.

Does this help you? Do you need some more data? Feel free to ask!



rt61pci (original (unpatched) from OpenSuSE 11.3)
==============================================

download
0,7 MBit/s

upstream
error (interrupted system call)


If you compare ndiswrapper with rt61pci patched, there is a difference
of about 6 MBits/s. The unpatched version can't be used at all.

Ok, so either the txpower handling in rt61pci needs to be reviewed or your
eeprom contents are crippled up. Not sure though ...

Is there a way to check this? Can I do anything to test?


Kind regards,
Andreas
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/