[Olsr-users] Freifunk´s ethernet-performace bottle neck

[h2o-post]

Hi Freifunker,

We are right befor of having a proper Freifunk based 5GHz turbo-a mesh
firmware in place. The tide about madwifi under Freifunk has turned in the
last weeks and adhoc meshes running in 5GHz are possible. Thanks to all
programmers involved in this.
The big "BUT" comes up with the network performance of our firmware. Applied
to the Atheros turbo-a in adhoc mode it leads currently to a working
wireless part, which can handle around 60 Bit/sek of troughput, but this
performance is not achiveable on our most deployed embedded hardware like
the ASUS wl-500gP. On such units, the ksoftirqd sofware irq handler takes to
much cpu power to achive this possible throughput, even while transfering
packets from one wired interface to antoher (like from WAN to LAN). Other
router formwares like Tomato or DD-WRT are able to handle this high
throughput.

Long story short: Looking at this issue I figured out, that the
ksoftirqd_CPU0 process in our firmware takes the most CPU power. While
searching for possible troubleshooting steps, I found a possible reason,
maybe. Have a look at
 <http://wl700g.homelinux.net/drupal/?q=comment/reply/110>
http://wl700g.homelinux.net/drupal/?q=comment/reply/110 ... they experienced
a
GPIO INTERRUPT STORM comming from a wrong configured interrrupt source.

Could that be the same problem on our side ? Could somebody imaging of what
gpio source could trigger that hugh among of software interrupts ?

Hopefully we solve this to be able of having a high speed Freifunk mesh ....


Bye4now from Bluse



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.olsr.org/pipermail/olsr-users/attachments/20080731/220c547e/attachment.html>