All of lore.kernel.org
 help / color / mirror / Atom feed
* [ath9k-devel] ath9k performance issues
@ 2011-04-05 15:44 Bernhard Walle
  2011-04-05 15:59 ` Felix Fietkau
  0 siblings, 1 reply; 15+ messages in thread
From: Bernhard Walle @ 2011-04-05 15:44 UTC (permalink / raw)
  To: ath9k-devel

Hello,

we're using a PCIe AR9280 card on a Octeon MIPS CPU. The system 
that does the transmission has a high CPU utilization:

iperf in UDP mode with 50M bandwith has ~90 % CPU utilization with a
high softirq proportion.

Using an Intel card (5300) has a lower CPU utilization (~60 %), but it's
no option to use it because the Linux driver has no access point mode.

Are there any tweaks to decrease the CPU utilization with the ath9k
driver? Any recommendations for the application to send the network
packets more efficient?


Regards,
Bernhard

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 15:44 [ath9k-devel] ath9k performance issues Bernhard Walle
@ 2011-04-05 15:59 ` Felix Fietkau
  2011-04-05 16:03   ` Serene Gud
  2011-04-05 16:19   ` Bernhard Walle
  0 siblings, 2 replies; 15+ messages in thread
From: Felix Fietkau @ 2011-04-05 15:59 UTC (permalink / raw)
  To: ath9k-devel

On 2011-04-05 5:44 PM, Bernhard Walle wrote:
> Hello,
>
> we're using a PCIe AR9280 card on a Octeon MIPS CPU. The system
> that does the transmission has a high CPU utilization:
>
> iperf in UDP mode with 50M bandwith has ~90 % CPU utilization with a
> high softirq proportion.
>
> Using an Intel card (5300) has a lower CPU utilization (~60 %), but it's
> no option to use it because the Linux driver has no access point mode.
>
> Are there any tweaks to decrease the CPU utilization with the ath9k
> driver? Any recommendations for the application to send the network
> packets more efficient?
What kernel version are you using? Are you using the in-kernel ath9k or 
are you using compat-wireless? I have added quite a few performance 
optimizations over time, as I'm working with much slower MIPS CPUs than 
the octeon.
In my tests, a 300 MHz MIPS 24K CPU is good enough for handling around 
90 mbit/s of traffic with only around 70% CPU utilization.

- Felix

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 15:59 ` Felix Fietkau
@ 2011-04-05 16:03   ` Serene Gud
  2011-04-05 16:32     ` Felix Fietkau
  2011-04-05 16:19   ` Bernhard Walle
  1 sibling, 1 reply; 15+ messages in thread
From: Serene Gud @ 2011-04-05 16:03 UTC (permalink / raw)
  To: ath9k-devel



--- On Tue, 4/5/11, Felix Fietkau <nbd@openwrt.org> wrote:

From: Felix Fietkau <nbd@openwrt.org>
Subject: Re: [ath9k-devel] ath9k performance issues
To: "Bernhard Walle" <bernhard@bwalle.de>
Cc: ath9k-devel at lists.ath9k.org
Date: Tuesday, April 5, 2011, 9:29 PM

On 2011-04-05 5:44 PM, Bernhard Walle wrote:
> Hello,
>
> we're using a PCIe AR9280 card on a Octeon MIPS CPU. The system
> that does the transmission has a high CPU utilization:
>
> iperf in UDP mode with 50M bandwith has ~90 % CPU utilization with a
> high softirq proportion.
>
> Using an Intel card (5300) has a lower CPU utilization (~60 %), but it's
> no option to use it because the Linux driver has no access point mode.
>
> Are there any tweaks to decrease the CPU utilization with the ath9k
> driver? Any recommendations for the application to send the network
> packets more efficient?
What kernel version are you using? Are you using the in-kernel ath9k or 
are you using compat-wireless? I have added quite a few performance 
optimizations over time, as I'm working with much slower MIPS CPUs than 
the octeon.
In my tests, a 300 MHz MIPS 24K CPU is good enough for handling around 
90 mbit/s of traffic with only around 70% CPU utilization.

- Felix


Now thats interesting Felix :) , how have you managed to get all these?

--SGS


_______________________________________________
ath9k-devel mailing list
ath9k-devel at lists.ath9k.org
https://lists.ath9k.org/mailman/listinfo/ath9k-devel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.ath9k.org/pipermail/ath9k-devel/attachments/20110405/eb912110/attachment.htm 

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 15:59 ` Felix Fietkau
  2011-04-05 16:03   ` Serene Gud
@ 2011-04-05 16:19   ` Bernhard Walle
  2011-04-05 16:54     ` Felix Fietkau
  1 sibling, 1 reply; 15+ messages in thread
From: Bernhard Walle @ 2011-04-05 16:19 UTC (permalink / raw)
  To: ath9k-devel

Hello Felix,

* Felix Fietkau <nbd@openwrt.org> [2011-04-05 17:59]:
> On 2011-04-05 5:44 PM, Bernhard Walle wrote:
> >
> >we're using a PCIe AR9280 card on a Octeon MIPS CPU. The system
> >that does the transmission has a high CPU utilization:
> >
> >iperf in UDP mode with 50M bandwith has ~90 % CPU utilization with a
> >high softirq proportion.
> >
> >Using an Intel card (5300) has a lower CPU utilization (~60 %), but it's
> >no option to use it because the Linux driver has no access point mode.
> >
> >Are there any tweaks to decrease the CPU utilization with the ath9k
> >driver? Any recommendations for the application to send the network
> >packets more efficient?
> What kernel version are you using? Are you using the in-kernel ath9k
> or are you using compat-wireless? I have added quite a few

I'm using 2.6.38.1 ath9k which is at ~80 % CPU, starting from 2.6.37.1
with ~90 % CPU which was still an improvement. Do you think that
compat-wireless gives still a performance boost?

> Which 
> performance optimizations over time, as I'm working with much slower
> MIPS CPUs than the octeon.
> In my tests, a 300 MHz MIPS 24K CPU is good enough for handling
> around 90 mbit/s of traffic with only around 70% CPU utilization.

That sounds interesting. Any other optimisations other than using
compat-wireless?


Regards,
Bernhard

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 16:03   ` Serene Gud
@ 2011-04-05 16:32     ` Felix Fietkau
  2011-04-05 16:54       ` Serene Gud
  0 siblings, 1 reply; 15+ messages in thread
From: Felix Fietkau @ 2011-04-05 16:32 UTC (permalink / raw)
  To: ath9k-devel

On 2011-04-05 6:03 PM, Serene Gud wrote:
> --- On *Tue, 4/5/11, Felix Fietkau /<nbd@openwrt.org>/* wrote:
>     From: Felix Fietkau <nbd@openwrt.org>
>     Subject: Re: [ath9k-devel] ath9k performance issues
>     To: "Bernhard Walle" <bernhard@bwalle.de>
>     Cc: ath9k-devel at lists.ath9k.org
>     Date: Tuesday, April 5, 2011, 9:29 PM
>
>     On 2011-04-05 5:44 PM, Bernhard Walle wrote:
>      > Hello,
>      >
>      > we're using a PCIe AR9280 card on a Octeon MIPS CPU. The system
>      > that does the transmission has a high CPU utilization:
>      >
>      > iperf in UDP mode with 50M bandwith has ~90 % CPU utilization with a
>      > high softirq proportion.
>      >
>      > Using an Intel card (5300) has a lower CPU utilization (~60 %),
>     but it's
>      > no option to use it because the Linux driver has no access point
>     mode.
>      >
>      > Are there any tweaks to decrease the CPU utilization with the ath9k
>      > driver? Any recommendations for the application to send the network
>      > packets more efficient?
>     What kernel version are you using? Are you using the in-kernel ath9k or
>     are you using compat-wireless? I have added quite a few performance
>     optimizations over time, as I'm working with much slower MIPS CPUs than
>     the octeon.
>     In my tests, a 300 MHz MIPS 24K CPU is good enough for handling around
>     90 mbit/s of traffic with only around 70% CPU utilization.
>
>     - Felix
>
>
>     Now thats interesting Felix :) , how have you managed to get all these?
What do you mean with 'get all these'?

- Felix

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 16:32     ` Felix Fietkau
@ 2011-04-05 16:54       ` Serene Gud
  2011-04-05 17:02         ` Felix Fietkau
  0 siblings, 1 reply; 15+ messages in thread
From: Serene Gud @ 2011-04-05 16:54 UTC (permalink / raw)
  To: ath9k-devel



On 2011-04-05 6:03 PM, Serene Gud wrote:

>
>? ???On 2011-04-05 5:44 PM, Bernhard Walle wrote:
>? ? ? > Hello,
>? ? ? >
>? ? ? > we're using a PCIe AR9280 card on a Octeon MIPS CPU. The system
>? ? ? > that does the transmission has a high CPU utilization:
>? ? ? >
>? ? ? > iperf in UDP mode with 50M bandwith has ~90 % CPU utilization with a
>? ? ? > high softirq proportion.
>? ? ? >
>? ? ? > Using an Intel card (5300) has a lower CPU utilization (~60 %),
>? ???but it's
>? ? ? > no option to use it because the Linux driver has no access point
>? ???mode.
>? ? ? >
>? ? ? > Are there any tweaks to decrease the CPU utilization with the ath9k
>? ? ? > driver? Any recommendations for the application to send the network
>? ? ? > packets more efficient?
>? ???What kernel version are you using? Are you using the in-kernel ath9k or
>? ???are you using compat-wireless? I have added quite a few performance
>? ???optimizations over time, as I'm working with much slower MIPS CPUs than
>? ???the octeon.
>? ???In my tests, a 300 MHz MIPS 24K CPU is good enough for handling around
>? ???90 mbit/s of traffic with only around 70% CPU utilization.
>
>? ???- Felix
>
>
>? ???Now thats interesting Felix :) , how have you managed to get all these?
What do you mean with 'get all these'?

- Felix


Sorry, I am new to these things; what I wanted to ask you was that with the kind of CPU info that you have; how were you able to get 90 mbit/s of traffic. I am using a much faster CPU with 1.66 Ghz 512K still I usually get 65mbit/s. I am using in-kernel ath9K. Are you using compat-wireless? Are there any other changes that you have made to your system? 
I have no idea if we can make the tx bitrate stable? If you have done that, kindly share.

Thanks!
SGS

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.ath9k.org/pipermail/ath9k-devel/attachments/20110405/980d21a1/attachment-0001.htm 

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 16:19   ` Bernhard Walle
@ 2011-04-05 16:54     ` Felix Fietkau
  2011-04-05 19:57       ` Bernhard Walle
  0 siblings, 1 reply; 15+ messages in thread
From: Felix Fietkau @ 2011-04-05 16:54 UTC (permalink / raw)
  To: ath9k-devel

On 2011-04-05 6:19 PM, Bernhard Walle wrote:
> Hello Felix,
>
> * Felix Fietkau<nbd@openwrt.org>  [2011-04-05 17:59]:
>>  On 2011-04-05 5:44 PM, Bernhard Walle wrote:
>>  >
>>  >we're using a PCIe AR9280 card on a Octeon MIPS CPU. The system
>>  >that does the transmission has a high CPU utilization:
>>  >
>>  >iperf in UDP mode with 50M bandwith has ~90 % CPU utilization with a
>>  >high softirq proportion.
>>  >
>>  >Using an Intel card (5300) has a lower CPU utilization (~60 %), but it's
>>  >no option to use it because the Linux driver has no access point mode.
>>  >
>>  >Are there any tweaks to decrease the CPU utilization with the ath9k
>>  >driver? Any recommendations for the application to send the network
>>  >packets more efficient?
>>  What kernel version are you using? Are you using the in-kernel ath9k
>>  or are you using compat-wireless? I have added quite a few
>
> I'm using 2.6.38.1 ath9k which is at ~80 % CPU, starting from 2.6.37.1
> with ~90 % CPU which was still an improvement. Do you think that
> compat-wireless gives still a performance boost?
Not sure. These numbers sound awfully high to me. Maybe you should do 
some profiling to figure out what's using all this CPU time.

>>  Which
>>  performance optimizations over time, as I'm working with much slower
>>  MIPS CPUs than the octeon.
>>  In my tests, a 300 MHz MIPS 24K CPU is good enough for handling
>>  around 90 mbit/s of traffic with only around 70% CPU utilization.
>
> That sounds interesting. Any other optimisations other than using
> compat-wireless?
No. I've pushed all of my enhancements upstream.

- Felix

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 16:54       ` Serene Gud
@ 2011-04-05 17:02         ` Felix Fietkau
  2011-04-05 17:17           ` Serene Gud
  2011-04-05 17:52           ` Kucherenko Valeriy
  0 siblings, 2 replies; 15+ messages in thread
From: Felix Fietkau @ 2011-04-05 17:02 UTC (permalink / raw)
  To: ath9k-devel

On 2011-04-05 6:54 PM, Serene Gud wrote:
>
>
>     On 2011-04-05 6:03 PM, Serene Gud wrote:
>
>      >
>      > On 2011-04-05 5:44 PM, Bernhard Walle wrote:
>      > > Hello,
>      > >
>      > > we're using a PCIe AR9280 card on a Octeon MIPS CPU. The system
>      > > that does the transmission has a high CPU utilization:
>      > >
>      > > iperf in UDP mode with 50M bandwith has ~90 % CPU utilization
>     with a
>      > > high softirq proportion.
>      > >
>      > > Using an Intel card (5300) has a lower CPU utilization (~60 %),
>      > but it's
>      > > no option to use it because the Linux driver has no access point
>      > mode.
>      > >
>      > > Are there any tweaks to decrease the CPU utilization with the ath9k
>      > > driver? Any recommendations for the application to send the network
>      > > packets more efficient?
>      > What kernel version are you using? Are you using the in-kernel
>     ath9k or
>      > are you using compat-wireless? I have added quite a few performance
>      > optimizations over time, as I'm working with much slower MIPS
>     CPUs than
>      > the octeon.
>      > In my tests, a 300 MHz MIPS 24K CPU is good enough for handling
>     around
>      > 90 mbit/s of traffic with only around 70% CPU utilization.
>      >
>      > - Felix
>      >
>      >
>      > Now thats interesting Felix :) , how have you managed to get all
>     these?
>     What do you mean with 'get all these'?
>
>     - Felix
>
>
>     Sorry, I am new to these things; what I wanted to ask you was that
>     with the kind of CPU info that you have; how were you able to get 90
>     mbit/s of traffic. I am using a much faster CPU with 1.66 Ghz 512K
>     still I usually get 65mbit/s. I am using in-kernel ath9K. Are you
>     using compat-wireless? Are there any other changes that you have
>     made to your system?
The device I used was using an AR7130 MIPS CPU. I set it up as an access 
point (with 4addr/wds mode enabled), then used another device as a 
4-addr client. Both were set up for bridging wireless to wired.
After that I ran iperf TCP sessions between two laptops (each connected 
to one of the embedded devices).

>     I have no idea if we can make the tx bitrate stable? If you have
>     done that, kindly share.
Are you using the ath9k rate control or are you using minstrel_ht? My 
tests were done with minstrel_ht. You can enable it by disabling the 
config option for the ath9k rate control.

- Felix

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 17:02         ` Felix Fietkau
@ 2011-04-05 17:17           ` Serene Gud
  2011-04-05 17:52           ` Kucherenko Valeriy
  1 sibling, 0 replies; 15+ messages in thread
From: Serene Gud @ 2011-04-05 17:17 UTC (permalink / raw)
  To: ath9k-devel



--- On Tue, 4/5/11, Felix Fietkau <nbd@openwrt.org> wrote:
On 2011-04-05 6:54 PM, Serene Gud wrote:
>
>
>? ???On 2011-04-05 6:03 PM, Serene Gud wrote:
>
>? ? ? >
>? ? ? > On 2011-04-05 5:44 PM, Bernhard Walle wrote:
>? ? ? > > Hello,
>? ? ? > >
>? ? ? > > we're using a PCIe AR9280 card on a Octeon MIPS CPU. The system
>? ? ? > > that does the transmission has a high CPU utilization:
>? ? ? > >
>? ? ? > > iperf in UDP mode with 50M bandwith has ~90 % CPU utilization
>? ???with a
>? ? ? > > high softirq proportion.
>? ? ? > >
>? ? ? > > Using an Intel card (5300) has a lower CPU utilization (~60 %),
>? ? ? > but it's
>? ? ? > > no option to use it because the Linux driver has no access point
>? ? ? > mode.
>? ? ? > >
>? ? ? > > Are there any tweaks to decrease the CPU utilization with the ath9k
>? ? ? > > driver? Any recommendations for the application to send the network
>? ? ? > > packets more efficient?
>? ? ? > What kernel version are you using? Are you using the in-kernel
>? ???ath9k or
>? ? ? > are you using compat-wireless? I have added quite a few performance
>? ? ? > optimizations over time, as I'm working with much slower MIPS
>? ???CPUs than
>? ? ? > the octeon.
>? ? ? > In my tests, a 300 MHz MIPS 24K CPU is good enough for handling
>? ???around
>? ? ? > 90 mbit/s of traffic with only around 70% CPU utilization.
>? ? ? >
>? ? ? > - Felix
>? ? ? >
>? ? ? >
>? ? ? > Now thats interesting Felix :) , how have you managed to get all
>? ???these?
>? ???What do you mean with 'get all these'?
>
>? ???- Felix
>
>
>? ???Sorry, I am new to these things; what I wanted to ask you was that
>? ???with the kind of CPU info that you have; how were you able to get 90
>? ???mbit/s of traffic. I am using a much faster CPU with 1.66 Ghz 512K
>? ???still I usually get 65mbit/s. I am using in-kernel ath9K. Are you
>? ???using compat-wireless? Are there any other changes that you have
>? ???made to your system?
The device I used was using an AR7130 MIPS CPU. I set it up as an access 
point (with 4addr/wds mode enabled), then used another device as a 
4-addr client. Both were set up for bridging wireless to wired.
After that I ran iperf TCP sessions between two laptops (each connected 
to one of the embedded devices).

Thanks Felix. I set up an Access point with master mode enables (with the help of hostapd). 

>? ???I have no idea if we can make the tx bitrate stable? If you have
>? ???done that, kindly share.
Are you using the ath9k rate control or are you using minstrel_ht? My 
tests were done with minstrel_ht. You can enable it by disabling the 
config option for the ath9k rate control.

My client had an intel card which uses it own rate control so I am using a USB wireless card now which uses minstrel but my Access point on the other hand uses ath9k_rate_control. I guess I should make them both use the same rate control algo. 

--SGS

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.ath9k.org/pipermail/ath9k-devel/attachments/20110405/003ee767/attachment.htm 

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 17:02         ` Felix Fietkau
  2011-04-05 17:17           ` Serene Gud
@ 2011-04-05 17:52           ` Kucherenko Valeriy
  2011-04-05 18:29             ` Felix Fietkau
  1 sibling, 1 reply; 15+ messages in thread
From: Kucherenko Valeriy @ 2011-04-05 17:52 UTC (permalink / raw)
  To: ath9k-devel

> Are you using the ath9k rate control or are you using minstrel_ht? My 
> tests were done with minstrel_ht. You can enable it by disabling the 
> config option for the ath9k rate control.
> 
> - Felix


This is interesting. I have a problem with rate control (I guess) too. Can you give me a clue on how to disable rate control for ath9k? What is a config option for ath9k? Where is it? Can I disable it by writing some value into /sys or maybe it's possible via modprobe ath9k --some_params?

Thanks.

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 17:52           ` Kucherenko Valeriy
@ 2011-04-05 18:29             ` Felix Fietkau
  2011-04-05 18:49               ` Larry Vaden
  0 siblings, 1 reply; 15+ messages in thread
From: Felix Fietkau @ 2011-04-05 18:29 UTC (permalink / raw)
  To: ath9k-devel

On 2011-04-05 7:52 PM, Kucherenko Valeriy wrote:
>>  Are you using the ath9k rate control or are you using minstrel_ht? My
>>  tests were done with minstrel_ht. You can enable it by disabling the
>>  config option for the ath9k rate control.
>>
>>  - Felix
>
>
> This is interesting. I have a problem with rate control (I guess)
> too.  Can you give me a clue on how to disable rate control for ath9k? What is
> a config option for ath9k? Where is it? Can I disable it by writing some
> value into /sys or maybe it's possible via modprobe ath9k --some_params?
In the kernel tree (.config) and in compat-wireless (config.mk) it's the 
option CONFIG_ATH9K_RATE_CONTROL.

- Felix

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 18:29             ` Felix Fietkau
@ 2011-04-05 18:49               ` Larry Vaden
  2011-04-05 18:57                 ` Larry Vaden
  0 siblings, 1 reply; 15+ messages in thread
From: Larry Vaden @ 2011-04-05 18:49 UTC (permalink / raw)
  To: ath9k-devel

On Tue, Apr 5, 2011 at 1:29 PM, Felix Fietkau <nbd@openwrt.org> wrote:
> On 2011-04-05 7:52 PM, Kucherenko Valeriy wrote:
>>> ?Are you using the ath9k rate control or are you using minstrel_ht? My
>>> ?tests were done with minstrel_ht. You can enable it by disabling the
>>> ?config option for the ath9k rate control.

HTH;  if not, please excuse the waste of time;  this is from stock
r26460 on Ubiquiti M900 and Loco M900:

root at OpenWrt:~# iw wlan0 station dump
Station 00:15:6d:9c:16:df (on wlan0)
        inactive time:  840 ms
        rx bytes:       628
        rx packets:     22
        tx bytes:       560
        tx packets:     11
        tx retries:     65
        tx failed:      7
        signal:         -97 dBm
        signal avg:     -96 dBm
        tx bitrate:     104.0 MBit/s MCS 13
        rx bitrate:     6.5 MBit/s MCS 0

best regards/ldv

-- 
Larry Vaden, CoFounder
Internet Texoma, Inc.
Serving Rural Texomaland Since 1995
We Care About Your Connection!

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 18:49               ` Larry Vaden
@ 2011-04-05 18:57                 ` Larry Vaden
  0 siblings, 0 replies; 15+ messages in thread
From: Larry Vaden @ 2011-04-05 18:57 UTC (permalink / raw)
  To: ath9k-devel

Again, HTH:  this is from a Bullet M2 running stock r26358:

Station 00:15:6d:f0:47:dd (on wlan0)
        inactive time:  44790 ms
        rx bytes:       10380739
        rx packets:     55088
        tx bytes:       96278984
        tx packets:     81367
        tx retries:     51061
        tx failed:      0
        signal:         -79 dBm
        signal avg:     -78 dBm
        tx bitrate:     48.0 MBit/s
        rx bitrate:     9.0 MBit/s

best regards/ldv
-- 
Larry Vaden, CoFounder
Internet Texoma, Inc.
Serving Rural Texomaland Since 1995
We Care About Your Connection!

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 16:54     ` Felix Fietkau
@ 2011-04-05 19:57       ` Bernhard Walle
  2011-04-05 20:09         ` Felix Fietkau
  0 siblings, 1 reply; 15+ messages in thread
From: Bernhard Walle @ 2011-04-05 19:57 UTC (permalink / raw)
  To: ath9k-devel

* Felix Fietkau <nbd@openwrt.org> [2011-04-05 18:54]:
> >
> >I'm using 2.6.38.1 ath9k which is at ~80 % CPU, starting from 2.6.37.1
> >with ~90 % CPU which was still an improvement. Do you think that
> >compat-wireless gives still a performance boost?
> Not sure. These numbers sound awfully high to me. Maybe you should
> do some profiling to figure out what's using all this CPU time.

Just to make sure: my Octeon runs at 300 MHz. Do you think that numbers
are too much? Userspace profiling showed that the CPU goes to sendto(),
of course. Which profiling method do you recommend for MIPS -- oprofile?


Regards,
Bernhard

^ permalink raw reply	[flat|nested] 15+ messages in thread

* [ath9k-devel] ath9k performance issues
  2011-04-05 19:57       ` Bernhard Walle
@ 2011-04-05 20:09         ` Felix Fietkau
  0 siblings, 0 replies; 15+ messages in thread
From: Felix Fietkau @ 2011-04-05 20:09 UTC (permalink / raw)
  To: ath9k-devel

On 2011-04-05 9:57 PM, Bernhard Walle wrote:
> * Felix Fietkau<nbd@openwrt.org>  [2011-04-05 18:54]:
>>  >
>>  >I'm using 2.6.38.1 ath9k which is at ~80 % CPU, starting from 2.6.37.1
>>  >with ~90 % CPU which was still an improvement. Do you think that
>>  >compat-wireless gives still a performance boost?
>>  Not sure. These numbers sound awfully high to me. Maybe you should
>>  do some profiling to figure out what's using all this CPU time.
>
> Just to make sure: my Octeon runs at 300 MHz. Do you think that numbers
> are too much? Userspace profiling showed that the CPU goes to sendto(),
> of course. Which profiling method do you recommend for MIPS -- oprofile?
Yes, I was using oprofile when I was working on improving ath9k performance.

- Felix

^ permalink raw reply	[flat|nested] 15+ messages in thread

end of thread, other threads:[~2011-04-05 20:09 UTC | newest]

Thread overview: 15+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-04-05 15:44 [ath9k-devel] ath9k performance issues Bernhard Walle
2011-04-05 15:59 ` Felix Fietkau
2011-04-05 16:03   ` Serene Gud
2011-04-05 16:32     ` Felix Fietkau
2011-04-05 16:54       ` Serene Gud
2011-04-05 17:02         ` Felix Fietkau
2011-04-05 17:17           ` Serene Gud
2011-04-05 17:52           ` Kucherenko Valeriy
2011-04-05 18:29             ` Felix Fietkau
2011-04-05 18:49               ` Larry Vaden
2011-04-05 18:57                 ` Larry Vaden
2011-04-05 16:19   ` Bernhard Walle
2011-04-05 16:54     ` Felix Fietkau
2011-04-05 19:57       ` Bernhard Walle
2011-04-05 20:09         ` Felix Fietkau

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.