[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[Mahesh Kumar]

Hello,

We have a critical issue in getting the high throughput, once we enable 
both the Aggregation (AMPDU) and Encryption (WPA) at the same time. (We 
are using the AR9160 chipset.)

Our test setup is like this,

[PC1] ---> [Access Point] --> [Client] --> [PC2].

We are sending a 100 Mbps of UDP traffic from PC1 to PC2. The 
Aggregation (AMPDU) & Encryption are enabled on both "Access point" and 
the "Client". Very low traffic ( < 1Mbps) is reaching at the PC2, that 
also occasionally. Most of the traffic is dropped at the Access point 
itself. When we check the transmit status descriptor@the Access 
point, it shows , lots of "Tx delimiter under-run error" & " Tx data 
under-run error".

Once the driver gets "under-run" error, it starts adjusting the Tx FIFO 
trigger level [TXCFG register, Frame Trigger level], and it finally 
reaches it's maximum. This register specifies the minimum number of 
bytes that must be DMAed  into the PCU TXFIFO, before the PCU initiates 
the frame on the air.

Any idea on why it's happening ?. Are we missing any special kind of 
configuration ?. Or is there any known issue when we enable both 
Aggregation (AMPDU) and encryption simultaneously ?.

The same setup is working, if we disable the Encryption ( getting around 
95.7 Mbps). Also the encryption alone is working, without the 
Aggregation, though we are getting a degraded performance ( max around 
30 Mbps.)
 

Thanks in advance,
Mahesh.

[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[Mahesh Kumar]

We are using the latest driver.
Kernel version is 2.6.13
armeb-linux-uclibc-gcc version 3.4.3, for xscale


charles at thewybles.com wrote:
> What version of the code?
> What version of the linux kernel?
> What compile tool chain?
> What distro? 
>
> Thanks. 
>
> Charles
>
>
> ------Original Message------
> From: Mahesh Kumar
> Sender: ath9k-devel-bounces at lists.ath9k.org
> To: ath9k-devel at lists.ath9k.org
> ReplyTo: mkumar at solectek.com
> Sent: Jul 28, 2008 4:11 PM
> Subject: [ath9k-devel] Atheros Throughput with Aggregation & Encryption
>
> Hello,
>
> We have a critical issue in getting the high throughput, once we enable 
> both the Aggregation (AMPDU) and Encryption (WPA) at the same time. (We 
> are using the AR9160 chipset.)
>
> Our test setup is like this,
>
> [PC1] ---> [Access Point] --> [Client] --> [PC2].
>
> We are sending a 100 Mbps of UDP traffic from PC1 to PC2. The 
> Aggregation (AMPDU) & Encryption are enabled on both "Access point" and 
> the "Client". Very low traffic ( < 1Mbps) is reaching@the PC2, that 
> also occasionally. Most of the traffic is dropped at the Access point 
> itself. When we check the transmit status descriptor at the Access 
> point, it shows , lots of "Tx delimiter under-run error" & " Tx data 
> under-run error".
>
> Once the driver gets "under-run" error, it starts adjusting the Tx FIFO 
> trigger level [TXCFG register, Frame Trigger level], and it finally 
> reaches it's maximum. This register specifies the minimum number of 
> bytes that must be DMAed  into the PCU TXFIFO, before the PCU initiates 
> the frame on the air.
>
> Any idea on why it's happening ?. Are we missing any special kind of 
> configuration ?. Or is there any known issue when we enable both 
> Aggregation (AMPDU) and encryption simultaneously ?.
>
> The same setup is working, if we disable the Encryption ( getting around 
> 95.7 Mbps). Also the encryption alone is working, without the 
> Aggregation, though we are getting a degraded performance ( max around 
> 30 Mbps.)
>  
>
> Thanks in advance,
> Mahesh.
> _______________________________________________
> ath9k-devel mailing list
> ath9k-devel at lists.ath9k.org
> https://lists.ath9k.org/mailman/listinfo/ath9k-devel
>
>
>
> Sent via BlackBerry from T-Mobile
> _______________________________________________
> ath9k-devel mailing list
> ath9k-devel at lists.ath9k.org
> https://lists.ath9k.org/mailman/listinfo/ath9k-devel
>
>   

[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[charles at thewybles.com]

What version of the code?
What version of the linux kernel?
What compile tool chain?
What distro? 

Thanks. 

Charles


------Original Message------
From: Mahesh Kumar
Sender: ath9k-devel-bounces at lists.ath9k.org
To: ath9k-devel at lists.ath9k.org
ReplyTo: mkumar at solectek.com
Sent: Jul 28, 2008 4:11 PM
Subject: [ath9k-devel] Atheros Throughput with Aggregation & Encryption

Hello,

We have a critical issue in getting the high throughput, once we enable 
both the Aggregation (AMPDU) and Encryption (WPA) at the same time. (We 
are using the AR9160 chipset.)

Our test setup is like this,

[PC1] ---> [Access Point] --> [Client] --> [PC2].

We are sending a 100 Mbps of UDP traffic from PC1 to PC2. The 
Aggregation (AMPDU) & Encryption are enabled on both "Access point" and 
the "Client". Very low traffic ( < 1Mbps) is reaching at the PC2, that 
also occasionally. Most of the traffic is dropped at the Access point 
itself. When we check the transmit status descriptor at the Access 
point, it shows , lots of "Tx delimiter under-run error" & " Tx data 
under-run error".

Once the driver gets "under-run" error, it starts adjusting the Tx FIFO 
trigger level [TXCFG register, Frame Trigger level], and it finally 
reaches it's maximum. This register specifies the minimum number of 
bytes that must be DMAed  into the PCU TXFIFO, before the PCU initiates 
the frame on the air.

Any idea on why it's happening ?. Are we missing any special kind of 
configuration ?. Or is there any known issue when we enable both 
Aggregation (AMPDU) and encryption simultaneously ?.

The same setup is working, if we disable the Encryption ( getting around 
95.7 Mbps). Also the encryption alone is working, without the 
Aggregation, though we are getting a degraded performance ( max around 
30 Mbps.)
 

Thanks in advance,
Mahesh.
_______________________________________________
ath9k-devel mailing list
ath9k-devel@lists.ath9k.org
https://lists.ath9k.org/mailman/listinfo/ath9k-devel



Sent via BlackBerry from T-Mobile

[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[Mahesh Kumar]

This happens with any traffic independent of UDP/TCP/.... But the 
problem starts when the Aggregation starts [with encryption = on].

Has anybody checked the Aggregation with Encryption = TKIP or AES. ?. 

Channelwidth = 40 MHz
Frequency = 5.785 GHz

Mahesh.

charles at thewybles.com wrote:
> Also does this occur with just UDP traffic? How about tcp? Sctp? Dccp? 
>
> What is your sample length? 5 minutes? 1 hour? 
>
> How about with another card? 
> Sure someone isn't running a microwave when your testing? :)
>
> Very interesting little problem you have going on. Look forward to seeing the resolution. 
>
> Charles Wyble
>
> ------Original Message------
> From: Mahesh Kumar
> Sender: ath9k-devel-bounces at lists.ath9k.org
> To: ath9k-devel at lists.ath9k.org
> ReplyTo: mkumar at solectek.com
> Sent: Jul 28, 2008 4:11 PM
> Subject: [ath9k-devel] Atheros Throughput with Aggregation & Encryption
>
> Hello,
>
> We have a critical issue in getting the high throughput, once we enable 
> both the Aggregation (AMPDU) and Encryption (WPA) at the same time. (We 
> are using the AR9160 chipset.)
>
> Our test setup is like this,
>
> [PC1] ---> [Access Point] --> [Client] --> [PC2].
>
> We are sending a 100 Mbps of UDP traffic from PC1 to PC2. The 
> Aggregation (AMPDU) & Encryption are enabled on both "Access point" and 
> the "Client". Very low traffic ( < 1Mbps) is reaching@the PC2, that 
> also occasionally. Most of the traffic is dropped at the Access point 
> itself. When we check the transmit status descriptor at the Access 
> point, it shows , lots of "Tx delimiter under-run error" & " Tx data 
> under-run error".
>
> Once the driver gets "under-run" error, it starts adjusting the Tx FIFO 
> trigger level [TXCFG register, Frame Trigger level], and it finally 
> reaches it's maximum. This register specifies the minimum number of 
> bytes that must be DMAed  into the PCU TXFIFO, before the PCU initiates 
> the frame on the air.
>
> Any idea on why it's happening ?. Are we missing any special kind of 
> configuration ?. Or is there any known issue when we enable both 
> Aggregation (AMPDU) and encryption simultaneously ?.
>
> The same setup is working, if we disable the Encryption ( getting around 
> 95.7 Mbps). Also the encryption alone is working, without the 
> Aggregation, though we are getting a degraded performance ( max around 
> 30 Mbps.)
>  
>
> Thanks in advance,
> Mahesh.
> _______________________________________________
> ath9k-devel mailing list
> ath9k-devel at lists.ath9k.org
> https://lists.ath9k.org/mailman/listinfo/ath9k-devel
>
>
>
> Sent via BlackBerry from T-Mobile

[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[charles at thewybles.com]

Also does this occur with just UDP traffic? How about tcp? Sctp? Dccp? 

What is your sample length? 5 minutes? 1 hour? 

How about with another card? 
Sure someone isn't running a microwave when your testing? :)

Very interesting little problem you have going on. Look forward to seeing the resolution. 

Charles Wyble

------Original Message------
From: Mahesh Kumar
Sender: ath9k-devel-bounces at lists.ath9k.org
To: ath9k-devel at lists.ath9k.org
ReplyTo: mkumar at solectek.com
Sent: Jul 28, 2008 4:11 PM
Subject: [ath9k-devel] Atheros Throughput with Aggregation & Encryption

Hello,

We have a critical issue in getting the high throughput, once we enable 
both the Aggregation (AMPDU) and Encryption (WPA) at the same time. (We 
are using the AR9160 chipset.)

Our test setup is like this,

[PC1] ---> [Access Point] --> [Client] --> [PC2].

We are sending a 100 Mbps of UDP traffic from PC1 to PC2. The 
Aggregation (AMPDU) & Encryption are enabled on both "Access point" and 
the "Client". Very low traffic ( < 1Mbps) is reaching at the PC2, that 
also occasionally. Most of the traffic is dropped at the Access point 
itself. When we check the transmit status descriptor at the Access 
point, it shows , lots of "Tx delimiter under-run error" & " Tx data 
under-run error".

Once the driver gets "under-run" error, it starts adjusting the Tx FIFO 
trigger level [TXCFG register, Frame Trigger level], and it finally 
reaches it's maximum. This register specifies the minimum number of 
bytes that must be DMAed  into the PCU TXFIFO, before the PCU initiates 
the frame on the air.

Any idea on why it's happening ?. Are we missing any special kind of 
configuration ?. Or is there any known issue when we enable both 
Aggregation (AMPDU) and encryption simultaneously ?.

The same setup is working, if we disable the Encryption ( getting around 
95.7 Mbps). Also the encryption alone is working, without the 
Aggregation, though we are getting a degraded performance ( max around 
30 Mbps.)
 

Thanks in advance,
Mahesh.
_______________________________________________
ath9k-devel mailing list
ath9k-devel@lists.ath9k.org
https://lists.ath9k.org/mailman/listinfo/ath9k-devel



Sent via BlackBerry from T-Mobile

[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[Luis R. Rodriguez]

On Mon, Jul 28, 2008 at 05:33:03PM -0700, Mahesh Kumar wrote:
> 
> We are using the latest driver.
> Kernel version is 2.6.13

Mahesh,

How were you able to do this? If you backported mac80211 down to 2.6.13
that is impressive as right now compat-wireless goes down to 2.6.21 (no
PCI yet tested) and on 2.6.22 you now get a kernel panic, which is why I
disabled 2.6.22 support.

Can you elaborate a little bit on this?

Can you consider moving to the wireless-testing kernel? That's what I
use for testing purposes.

  Luis

[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[Mahesh Kumar]

Thanks a lot Luis,

Can you please check what happens if both Aggregation & Encryption 
enabled ?.


Regards,
Mahesh.

Luis R. Rodriguez wrote:
> On Mon, Jul 28, 2008 at 05:33:03PM -0700, Mahesh Kumar wrote:
>> We are using the latest driver.
>> Kernel version is 2.6.13
>
> Mahesh,
>
> How were you able to do this? If you backported mac80211 down to 2.6.13
> that is impressive as right now compat-wireless goes down to 2.6.21 (no
> PCI yet tested) and on 2.6.22 you now get a kernel panic, which is why I
> disabled 2.6.22 support.
>
> Can you elaborate a little bit on this?
>
> Can you consider moving to the wireless-testing kernel? That's what I
> use for testing purposes.
>
>   Luis
>

[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[Mahesh Kumar]

Just an update,

I am using the latest Fusion Driver from Atheros. I compared the code 
with ath9k, and it's pretty much the same, than using different naming 
convention. So I assume the problem should be there with latest ath9k 
also, when we enable the Aggregation (AMPDU) and Encryption together. It 
would be great if somebody already did the same/similar test with ath9k, 
share the result.

Regarding the fusion driver,

It's not yet integrated to the kernel 2.6.13. It's built as modules.

Regards,
Mahesh.

Mahesh Kumar wrote:
>
> Thanks a lot Luis,
>
> Can you please check what happens if both Aggregation & Encryption 
> enabled ?.
>
>
> Regards,
> Mahesh.
>
> Luis R. Rodriguez wrote:
>> On Mon, Jul 28, 2008 at 05:33:03PM -0700, Mahesh Kumar wrote:
>>> We are using the latest driver.
>>> Kernel version is 2.6.13
>>
>> Mahesh,
>>
>> How were you able to do this? If you backported mac80211 down to 2.6.13
>> that is impressive as right now compat-wireless goes down to 2.6.21 (no
>> PCI yet tested) and on 2.6.22 you now get a kernel panic, which is why I
>> disabled 2.6.22 support.
>>
>> Can you elaborate a little bit on this?
>>
>> Can you consider moving to the wireless-testing kernel? That's what I
>> use for testing purposes.
>>
>>   Luis
>>
>
>

[ath9k-devel] Atheros Throughput with Aggregation & Encryption

[Sujith]

Mahesh Kumar wrote:
 > I am using the latest Fusion Driver from Atheros. I compared the code 
 > with ath9k, and it's pretty much the same, than using different naming 
 > convention. So I assume the problem should be there with latest ath9k 
 > also, when we enable the Aggregation (AMPDU) and Encryption together. It 
 > would be great if somebody already did the same/similar test with ath9k, 
 > share the result.
 > 
 > Regarding the fusion driver,
 > 
 > It's not yet integrated to the kernel 2.6.13. It's built as modules.

The Fusion driver was written for the net80211 stack and is a completely different beast
from ath9k. Please don't post issues that you find in Fusion in ath9k-devel.

Sujith
-- 
http://sujith-m.blogspot.com