[ath9k-devel] [PATCH] ath9k: Support RSN Mesh

[gtolon at inti.gob.ar <gtolon@inti.gob.ar>]

Hi Chun-Yeow.

El 02/12/2011 12:28 p.m., Yeoh Chun-Yeow escribi?:
> Hi, Gabriel
>
> nohwcrypt=1 will make the ath9k driver to use software crypto instead 
> of hardware crypto. My Atheros chipset is able to support hardware 
> encryption and decryption for CCMP and AES. Thus, when we setup the 
> 802.11s mesh network with security (SAE), the throughput for single 
> hop communication degrades significantly if we use software crypto. We 
> observe the HT throughput when we applied the patch to ensure unicast 
> frame is sent with hardware encryption. However, the throughput is 
> dropped for multi-hop communication. Possible due to the ath9k driver 
> since we have no such problem when security is disable.

Ok, i understand, so if it works well without security it has nothing to 
do with the problem i mentioned.

>
> Regarding the adhoc batman-adv, did you setup the adhoc mode using 
> security such as WPA2? What is the performance that you obtained?
>
> Regards,
> Chun-Yeow

We are not using security yet, because we are having problems with 
throughput tests without encryption, and it's difficult to find where 
the problem is.

Regards

Gabriel

>
>
> On Fri, Dec 2, 2011 at 9:26 PM, <gtolon@inti.gob.ar 
> <mailto:gtolon@inti.gob.ar>> wrote:
>
>     Hi Chun-Yeow
>
>     I don't know if this is usefull for you, but we are using ad hoc
>     with batman-adv, and we had some problems when the nodes didn't
>     choose good routes for high throughputs for relatively long
>     distances. Asking to the batman-adv list they explained us that
>     the algorithm to choose the routes is based on low rate
>     broadcasts, so the nodes could choose routes that with high rates
>     become bad. They suggested us to increase the multicast rates, so
>     that the nodes could "realize" if a route were not good for high
>     throughputs, and it worked better, the nodes chose routes with
>     more hops for long distances getting a better throughput than
>     using less hops with larger distances.
>
>     By the way, why with nohwcrypt=1 the throughput get worse?
>
>     Regards
>
>     Gabriel
>