Significiant performance differences between ath5k and ath9k in 802.11a

Significiant performance differences between ath5k and ath9k in 802.11a

[Joerg Pommnitz]

Hello all,
I have a strange performance problem with ath9k. I have two otherwise identical systems with an ath5k-supported card in system ATH5K and an ath9k supported card in system ATH9K.
Doing an UDP iperf run were ATH5K is the client/sender and ATH9K is the server/receiver I easily get data rates of 31MBit/s (tx data rate on the sender is manually set to 54 MBit/s).
Running the same test with ATH9K as client/sender and ATH5K as server/receiver gives maximum data rates of about 22MBit/s, about 30% less.

Doing the test against another ath9k system (e.g. ATH9K ==> ATH9K2) gives
the same max data rate of only 22 MBit/s, so the problem seems to be the
ath9k driver or the hardware. Can anybody confirm these problems? Is this
a known limitation? Any buttons to push to get things going faster?

Details:
the ath5k card is a AR2413 based design (Compex WLM54AG), the ath9k card
is a AR9220 (Compex WLM200NX 6A). The systems are connected in IBSS mode 
on channel 40 (5200MHz).
The kernel version is wireless-testing as of January, 26. The kernel is
slightly patched to enable IBSS in 802.11a (we are allowed to do this).

Can anybody shed some light on this strange issue? Luis?

-- 
Regards
       Joerg

__________________________________________________
Do You Yahoo!?
Sie sind Spam leid? Yahoo! Mail verfügt über einen herausragenden Schutz gegen Massenmails. 
http://mail.yahoo.com 

Re: [ath9k-devel] Significiant performance differences between ath5k and ath9k in 802.11a

[rootkit85]

On Thu, Jan 28, 2010 at 4:13 PM, Joerg Pommnitz <pommnitz@yahoo.com> wrote:
> Hello all,
> I have a strange performance problem with ath9k. I have two otherwise identical systems with an ath5k-supported card in system ATH5K and an ath9k supported card in system ATH9K.
> Doing an UDP iperf run were ATH5K is the client/sender and ATH9K is the server/receiver I easily get data rates of 31MBit/s (tx data rate on the sender is manually set to 54 MBit/s).
> Running the same test with ATH9K as client/sender and ATH5K as server/receiver gives maximum data rates of about 22MBit/s, about 30% less.
>
> Doing the test against another ath9k system (e.g. ATH9K ==> ATH9K2) gives
> the same max data rate of only 22 MBit/s, so the problem seems to be the
> ath9k driver or the hardware. Can anybody confirm these problems? Is this
> a known limitation? Any buttons to push to get things going faster?
>
> Details:
> the ath5k card is a AR2413 based design (Compex WLM54AG), the ath9k card
> is a AR9220 (Compex WLM200NX 6A). The systems are connected in IBSS mode
> on channel 40 (5200MHz).
> The kernel version is wireless-testing as of January, 26. The kernel is
> slightly patched to enable IBSS in 802.11a (we are allowed to do this).
>
> Can anybody shed some light on this strange issue? Luis?
>
> --
> Regards
>       Joerg
>
> __________________________________________________
> Do You Yahoo!?
> Sie sind Spam leid? Yahoo! Mail verfügt über einen herausragenden Schutz gegen Massenmails.
> http://mail.yahoo.com
> _______________________________________________
> ath9k-devel mailing list
> ath9k-devel@lists.ath9k.org
> https://lists.ath9k.org/mailman/listinfo/ath9k-devel
>

Can you try in AP-client mode? I think you'll get more throughput so.

Re: [ath9k-devel] Significiant performance differences between ath5k and ath9k in 802.11a

[Joerg Pommnitz]

--- rootkit85@yahoo.it <rootkit85@yahoo.it> wrote:

> 
> Can you try in AP-client mode? I think you'll get more
> throughput so.
> 

No, IBSS is what I'm interested in. And the point is, that there is a 30% performance difference between ath5k (and Madwifi) and ath9k. Even if the
performance in AP mode would be better, this does not change the problem
in IBSS mode.

Regards
  Joerg

__________________________________________________
Do You Yahoo!?
Sie sind Spam leid? Yahoo! Mail verfügt über einen herausragenden Schutz gegen Massenmails. 
http://mail.yahoo.com 

Re: [ath9k-devel] Significiant performance differences between ath5k and ath9k in 802.11a

[Felix Fietkau]

On 2010-01-29 9:10 AM, Joerg Pommnitz wrote:
> --- rootkit85@yahoo.it <rootkit85@yahoo.it> wrote:
> 
>> 
>> Can you try in AP-client mode? I think you'll get more
>> throughput so.
>> 
> 
> No, IBSS is what I'm interested in. And the point is, that there is
> a 30% performance difference between ath5k (and Madwifi) and ath9k. Even
> if the performance in AP mode would be better, this does not change the
> problem in IBSS mode.
I can confirm these numbers for AP/Client mode as well. I've even tested
legacy communication between ath9k ap and sta, producing the same
result. Additionally, I can rule out the rate control algorithm, as
using minstrel instead of the ath9k RC produces a very similar
throughput test result.

- Felix

Re: Significiant performance differences between ath5k and ath9k in 802.11a

[Adam Wozniak]

Your numbers for ath9k in the 5GHz band jibe with what I've observed for 
the ar9170 in ad-hoc mode.
If you go down to 2.4GHz you'll see about 16MBit/s.

I have no other numbers to offer for comparison.

--Adam

Joerg Pommnitz wrote:
> Hello all,
> I have a strange performance problem with ath9k. I have two otherwise identical systems with an ath5k-supported card in system ATH5K and an ath9k supported card in system ATH9K.
> Doing an UDP iperf run were ATH5K is the client/sender and ATH9K is the server/receiver I easily get data rates of 31MBit/s (tx data rate on the sender is manually set to 54 MBit/s).
> Running the same test with ATH9K as client/sender and ATH5K as server/receiver gives maximum data rates of about 22MBit/s, about 30% less.
>
> Doing the test against another ath9k system (e.g. ATH9K ==> ATH9K2) gives
> the same max data rate of only 22 MBit/s, so the problem seems to be the
> ath9k driver or the hardware. Can anybody confirm these problems? Is this
> a known limitation? Any buttons to push to get things going faster?
>
> Details:
> the ath5k card is a AR2413 based design (Compex WLM54AG), the ath9k card
> is a AR9220 (Compex WLM200NX 6A). The systems are connected in IBSS mode 
> on channel 40 (5200MHz).
> The kernel version is wireless-testing as of January, 26. The kernel is
> slightly patched to enable IBSS in 802.11a (we are allowed to do this).
>
> Can anybody shed some light on this strange issue? Luis?
>
>   

Re: [ath9k-devel] Significiant performance differences between ath5k and ath9k in 802.11a

[Joerg Pommnitz]

Luis,
with several people confirming the performance regression relative to AR5xxx/ath5k, can you tell us whether this will eventually be fixed? We are close to a purchase decision and would really like to upgrade to a recent chip set.
-- 
Regards
       Joerg


--- Felix Fietkau <nbd@openwrt.org> schrieb am Fr, 29.1.2010:

> Von: Felix Fietkau <nbd@openwrt.org>
> Betreff: Re: [ath9k-devel] Significiant performance differences between ath5k  and ath9k in 802.11a
> An: "Joerg Pommnitz" <pommnitz@yahoo.com>
> CC: rootkit85@yahoo.it, linux-wireless@vger.kernel.org, ath9k-devel@lists.ath9k.org, ath5k-devel@lists.ath5k.org
> Datum: Freitag, 29. Januar 2010, 9:23
> On 2010-01-29 9:10 AM, Joerg Pommnitz
> wrote:
> > --- rootkit85@yahoo.it
> <rootkit85@yahoo.it>
> wrote:
> > 
> >> 
> >> Can you try in AP-client mode? I think you'll get
> more
> >> throughput so.
> >> 
> > 
> > No, IBSS is what I'm interested in. And the point is,
> that there is
> > a 30% performance difference between ath5k (and
> Madwifi) and ath9k. Even
> > if the performance in AP mode would be better, this
> does not change the
> > problem in IBSS mode.
> I can confirm these numbers for AP/Client mode as well.
> I've even tested
> legacy communication between ath9k ap and sta, producing
> the same
> result. Additionally, I can rule out the rate control
> algorithm, as
> using minstrel instead of the ath9k RC produces a very
> similar
> throughput test result.
> 
> - Felix
> 

__________________________________________________
Do You Yahoo!?
Sie sind Spam leid? Yahoo! Mail verfügt über einen herausragenden Schutz gegen Massenmails. 
http://mail.yahoo.com