Modify

Opened 5 years ago

Closed 4 years ago

Last modified 4 years ago

#12364 closed defect (no_response)

ath9k performance worsens

Reported by: anonymous Owned by: developers
Priority: response-needed Milestone: Barrier Breaker 14.07
Component: packages Version: Trunk
Keywords: Cc:

Description

ath9k performance on a wzr hp g300nh dropped alot since sbk modifications in trunk recently, only around 35mbit with iperf reachable (throughput) on HT40, which is very bad. same location was able to deliver 50-55mbit before. in very early revisions even saw above 60mbit

Attachments (1)

ath9k-Fix-reporting-of-RX-STBC-streams-to-userspace.patch (810 bytes) - added by Anton van Bohemen <avbohemen@…> 5 years ago.
patch for mac80211

Download all attachments as: .zip

Change History (24)

comment:1 Changed 5 years ago by jow

  • Resolution set to too_vague
  • Status changed from new to closed

comment:2 Changed 5 years ago by unwired <mauermann@…>

  • Resolution too_vague deleted
  • Status changed from closed to reopened

Same here on a TL-WDR3600. I barely get more than 25 Mbits on 20MHz channel. Original TP-Link firmware was more than 2 times faster under same test conditions. Seems to me that ath9k supports only 1 stream. At least this is what iw list reports:

root@ap:~# iw list
Wiphy phy0
	Band 1:
		Capabilities: 0x11ef
			RX LDPC
			HT20/HT40
			SM Power Save disabled
			RX HT20 SGI
			RX HT40 SGI
			TX STBC
			RX STBC 1-stream
			Max AMSDU length: 3839 bytes
			DSSS/CCK HT40

If you ask google about "ath9k fix reporting of RX STBC streams to userspace" you'll find a patch from January 2011. However, I don't know if it's already in ath9k and if it would really improve performance.

comment:3 Changed 5 years ago by nbd

  • Priority changed from normal to response-needed

please try the current version

comment:4 Changed 5 years ago by anonymous

I have similar problems with a tl-wr2543nd.
In 11ng throughput is around 40Mbps if client is next to the router. A few meters away it drops to a horrible 0,5Mbps tand the router starts complaining that he received too few ACKs from client.
With 11g throughput is around 8Mbps in the same environment conditions.
With stock firmware I get a stable 10Mbps

With a WL500gP and ath9k card I don't have these problems and throughput is around 8Mbps in 11ng (max 40Mbps in best conditions ).
I'm using AArc1 here.

comment:5 Changed 5 years ago by nbd

please try current trunk

comment:6 Changed 5 years ago by mauermann@…

Just tried trunk r35938. No change in performance compared to AA :-(
iw list gives same output as before, so it's still working with only 1 stream.

comment:7 Changed 5 years ago by nbd

The iw list output does not mean that it's working with only one stream. It means that STBC is only supported for single-stream rates, which is normal.

comment:8 Changed 5 years ago by nbd

Please show me the throughput for both directions separately. And while running a throughput test in the AP -> Client direction, show me the output of

cat /sys/kernel/debug/ieee80211/phy*/netdev:*/stations/*/rc_stats

comment:9 Changed 5 years ago by mauermann@…

AA rc1
iperf AP -> Client: 31.1 Mbits/sec, Client -> AP: 28.4 Mbits/sec

type      rate     throughput  ewma prob   this prob  this succ/attempt   success    attempts
HT20/LGI    MCS0        6.0       92.0      100.0          0(  0)         15          16
HT20/LGI    MCS1        9.6       75.0        0.0          0(  0)         17          18
HT20/LGI    MCS2       14.1       75.3      100.0          0(  0)         13          16
HT20/LGI    MCS3       18.3       75.0        0.0          0(  0)         14          15
HT20/LGI    MCS4       35.2      100.0      100.0          0(  0)         14          14
HT20/LGI    MCS5       43.3       97.0      100.0          0(  0)         15          17
HT20/LGI    MCS6       43.1       88.0      100.0          0(  0)        159         227
HT20/LGI    MCS7       43.2       79.6       50.0          0(  0)       1052        1743
HT20/SGI    MCS0        7.2       99.2      100.0          0(  0)         15          16
HT20/SGI    MCS1       12.2       85.9      100.0          0(  0)         18          19
HT20/SGI    MCS2       20.7      100.0      100.0          0(  0)         14          14
HT20/SGI    MCS3       26.9      100.0      100.0          0(  0)         14          14
HT20/SGI    MCS4       37.9       98.5      100.0          0(  0)        255         294
HT20/SGI    MCS5       43.1       88.8      100.0          0(  0)       2325        2847
HT20/SGI  t MCS6       47.5       89.3       90.0        109(121)       3234        4306
HT20/SGI T PMCS7       48.8       82.9       83.5         66( 79)      19340       23864

Trunk
iperf AP -> Client: 25.3 Mbits/sec, Client -> AP: 35.4 Mbits/sec

type         rate     throughput  ewma prob   this prob  retry   this succ/attempt   success    attempts
CCK/LP        1.0M           0.6       74.9         0.0      2              0(  0)        46          54
CCK/SP        2.0M           1.6       90.4       100.0      2              0(  0)        18          20
CCK/SP        5.5M           3.9       94.1       100.0      2              0(  0)        16          19
CCK/SP       11.0M           6.7       97.4       100.0      2              0(  0)        18          19
HT20/LGI     MCS0            5.1       79.5       100.0      2              0(  0)        14          18
HT20/LGI     MCS1           12.8      100.0       100.0      2              0(  0)        19          19
HT20/LGI     MCS2           17.3       93.3       100.0      3              0(  0)        15          17
HT20/LGI     MCS3           18.3       75.0         0.0      4              0(  0)        19          20
HT20/LGI     MCS4           31.8       90.7       100.0      4              0(  0)        14          17
HT20/LGI  t  MCS5           43.1       96.8       100.0      5              0(  0)       741         820
HT20/LGI     MCS6           30.2       61.7         0.0      5              0(  1)       372         449
HT20/LGI T P MCS7           49.1       90.6        89.8      5            169(188)     33542       38115
HT20/SGI     MCS0            6.4       89.6       100.0      2              0(  0)        15          18
HT20/SGI     MCS1           13.5       96.0       100.0      2              0(  0)        18          20
HT20/SGI     MCS2           20.2       98.1       100.0      3              0(  0)        16          17
HT20/SGI     MCS3           26.8      100.0       100.0      4              0(  0)        20          20
HT20/SGI     MCS4           33.7       87.9       100.0      4              0(  0)        21          26
HT20/SGI     MCS5           15.5       32.3         0.0      5              0(  1)       132         158
HT20/SGI     MCS6           42.7       80.5       100.0      5              1(  1)       158         325
HT20/SGI     MCS7           12.0       20.6         0.0      5              0(  1)       455         995

comment:10 Changed 5 years ago by Anton van Bohemen <avbohemen@…>

If you ask google about "ath9k fix reporting of RX STBC streams to userspace" you'll find a patch from January 2011. However, I don't know if it's already in ath9k and if it would really improve performance.

The patch you speak of is ath9k: Fix reporting of RX STBC streams to userspace. It is not yet taken in the current mac80211 package. I adapted the patch and did a test. Indeed, iw list now shows:

Wiphy phy0
        Band 1:
                Capabilities: 0x12ce
                        HT20/HT40
                        SM Power Save disabled
                        RX HT40 SGI
                        TX STBC
                        RX STBC 2-streams
                        Max AMSDU length: 3839 bytes
                        DSSS/CCK HT40
                Maximum RX AMPDU length 65535 bytes (exponent: 0x003)
                Minimum RX AMPDU time spacing: 8 usec (0x06)
                HT TX/RX MCS rate indexes supported: 0-15

But I did not see much difference in throughput.
Without the patch (r35873):

type         rate     throughput  ewma prob   this prob  retry   this succ/attempt   success    attempts
CCK/LP        1.0M           0.9       99.9       100.0      2              0(  0)       166         171
CCK/SP        2.0M           1.8       99.8       100.0      2              0(  0)       115         122
CCK/SP        5.5M           4.2       99.9       100.0      2              0(  0)       113         120
CCK/SP       11.0M           6.8       99.9       100.0      2              0(  0)       118         123
HT20/LGI     MCS0            6.5       99.9       100.0      2              0(  0)       120         123
HT20/LGI     MCS1           12.9       99.9       100.0      2              0(  0)       122         131
HT20/LGI     MCS2           17.8       93.6       100.0      3              0(  0)       132         144
HT20/LGI  t  MCS3           24.7       99.0       100.0      3              0(  0)       134         141
HT20/LGI T P MCS4           34.1       94.0        97.8      4             45( 46)    120867      138641
HT20/LGI     MCS5            8.5       18.4         0.0      4              0(  0)     75957      101944
HT20/LGI     MCS6            0.0        0.0         0.0      1              0(  0)       459        3033
HT20/LGI     MCS7            0.0        0.0         0.0      1              0(  1)        10        3033

With the patch (r36033):

type         rate     throughput  ewma prob   this prob  retry   this succ/attempt   success    attempts
CCK/LP        1.0M           0.9       99.9       100.0      2              0(  0)        75          79
CCK/SP        2.0M           1.8       99.9       100.0      2              0(  0)        60          62
CCK/SP        5.5M           4.2       99.9       100.0      2              0(  0)        60          61
CCK/SP       11.0M           5.1       74.9         0.0      2              0(  1)        57          59
HT20/LGI     MCS0            5.9       89.3       100.0      2              0(  0)        56          59
HT20/LGI     MCS1           13.1       99.9       100.0      2              0(  0)        60          63
HT20/LGI     MCS2           17.8       92.5       100.0      2              0(  0)        81          91
HT20/LGI  t  MCS3           24.4       96.5       100.0      2              0(  0)       293         327
HT20/LGI T P MCS4           31.5       85.3        86.6      3            136(157)     93947      109025
HT20/LGI     MCS5           28.6       60.2       100.0      4              1(  1)     35698       56799
HT20/LGI     MCS6            0.0        0.0         0.0      1              0(  0)       198        1296
HT20/LGI     MCS7            0.0        0.0         0.0      1              0(  1)         1         997

By the way, my router is a Netgear WNDR3700v1, and I did not have any issues to begin with. Just wanted to know if the patch made a difference. For the patch to work, you need at least an Atheros 9280 rev 2 chip. My router has one, so it now reports 2 streams.

Changed 5 years ago by Anton van Bohemen <avbohemen@…>

patch for mac80211

comment:11 Changed 5 years ago by nbd

That patch is wrong. Reporting only one stream for STBC is correct, the hardware does not support 2-stream STBC.

comment:12 Changed 5 years ago by nbd

please try the latest version

comment:13 Changed 5 years ago by nbd

  • Resolution set to no_response
  • Status changed from reopened to closed

comment:14 Changed 5 years ago by mauermann@…

  • Resolution no_response deleted
  • Status changed from closed to reopened

Just tried latest trunk, but there is no change in performance. Still getting no more than 30...35 Mbits with iperf under optimal conditions. iw list still shows only RX STBC 1-stream. However, my hardware (TP-Link TL-WDR3600) supports 2 streams.

comment:15 Changed 5 years ago by nbd

Your hardware supports two streams, but not two-stream STBC.
Any details on what client you're using in trunk, what throughput you're expecting, and the contents of the rate control stats:

cat /sys/kernel/debug/ieee80211/phy*/netdev:*/stations/*/rc_stats

comment:16 Changed 5 years ago by mauermann@…

First day with stock firmware on my WDR3600 was very impressive. For example, I got more than 4 MB/s from a Samba server to a Nexus 7. After flashing Openwrt I was a little bit disappointed because throughput wasn't noticeable faster than my old 11g router. See my comment 2 above. With current trunk I'm now at 2 MB/s from the Samba server to the Nexus 7 under same conditions and iperf maxing out at 35 Mbits.

If it's not related to STBC, what else can explain the poor performance compared to stock firmware? Of course I used same channel settings (20 MHz because of neighboring WLANs) on stock firmware.

Here are again some rc_stats:

type         rate     throughput  ewma prob   this prob  retry   this succ/attempt   success    attempts
CCK/LP        1.0M           0.9       99.9       100.0      0              0(  0)       202         203
CCK/SP        2.0M           0.0        0.0         0.0      0              0(  0)         0           0
CCK/SP        5.5M           0.0        0.0         0.0      0              0(  0)         0           0
CCK/SP       11.0M           0.0        0.0         0.0      0              0(  0)         0           0
HT20/LGI     MCS0            6.5      100.0       100.0      1              0(  0)         1           1
HT20/LGI     MCS1           12.8      100.0       100.0      4              0(  0)         1           1
HT20/LGI     MCS2           18.7      100.0       100.0      0              0(  0)         1           1
HT20/LGI     MCS3           24.4      100.0       100.0      0              0(  0)         1           1
HT20/LGI     MCS4           35.0      100.0       100.0      0              0(  0)         1           1
HT20/LGI     MCS5           44.5      100.0       100.0      0              0(  0)         1           1
HT20/LGI     MCS6           46.8       95.8       100.0      5              0(  0)        10          16
HT20/LGI  t  MCS7           49.4       91.0        85.7      5              0(  0)      5565        6153
HT20/SGI     MCS0            7.1      100.0       100.0      0              0(  0)         1           1
HT20/SGI     MCS1           14.1      100.0       100.0      0              0(  0)         1           1
HT20/SGI     MCS2           20.5      100.0       100.0      0              0(  0)         1           1
HT20/SGI     MCS3           26.8      100.0       100.0      0              0(  0)         1           1
HT20/SGI     MCS4           38.4      100.0       100.0      5              0(  0)         2           2
HT20/SGI     MCS5           48.5      100.0       100.0      5              0(  0)         6           6
HT20/SGI     MCS6           48.2       90.9       100.0      5              0(  0)      1001        1117
HT20/SGI T P MCS7           52.3       89.0        85.1      5            183(215)     83747       92499


type         rate     throughput  ewma prob   this prob  retry   this succ/attempt   success    attempts
CCK/LP        1.0M           0.9       99.9       100.0      0              0(  0)       174         177
CCK/SP        2.0M           0.0        0.0         0.0      0              0(  0)         0           0
CCK/SP        5.5M           0.0        0.0         0.0      0              0(  0)         0           0
CCK/SP       11.0M           0.0        0.0         0.0      0              0(  0)         0           0
HT20/LGI     MCS0            6.5      100.0       100.0      1              0(  0)         1           1
HT20/LGI     MCS1           12.8      100.0       100.0      4              0(  0)         1           1
HT20/LGI     MCS2           18.7      100.0       100.0      0              0(  0)         1           1
HT20/LGI     MCS3           24.4      100.0       100.0      4              0(  0)        22          22
HT20/LGI     MCS4           34.9       99.4       100.0      4              0(  0)      1753        1969
HT20/LGI     MCS5           35.7       80.2       100.0      5              0(  0)      8046        9729
HT20/LGI  t  MCS6           45.3       92.7       100.0      5              0(  0)      3982        5214
HT20/LGI T P MCS7           48.9       90.1        87.1      5            142(163)     11938       13892

comment:17 Changed 5 years ago by nbd

wireless stats look clean. did you use samba on the device or is it running on another server in your lan?

comment:18 Changed 5 years ago by mauermann@…

Samba is running on another server connected to one of the 1Gb LAN ports

comment:19 Changed 5 years ago by nbd

please try latest trunk

comment:20 Changed 5 years ago by mauermann@…

no change with r37108

comment:21 Changed 4 years ago by nbd

please try latest again.

comment:22 Changed 4 years ago by nbd

  • Resolution set to no_response
  • Status changed from reopened to closed

comment:23 Changed 4 years ago by jow

  • Milestone changed from Attitude Adjustment 12.09 to Barrier Breaker 14.07

Milestone Attitude Adjustment 12.09 deleted

Add Comment

Modify Ticket

Action
as closed .
The resolution will be deleted. Next status will be 'reopened'.
Author


E-mail address and user name can be saved in the Preferences.

 
Note: See TracTickets for help on using tickets.