[Click] Click CPU load

[Torquato Bertani]

Hi Roberto,

a board like asus has a limited computational power to handle so many
pkt/s, here some reasons:

-roofnet routing is heavy to run
-click in userlevel is not very performant, kernel module is better
-madwifi driver are not polling so you have interrupt for each pkt
-you can't run an iperf session from the asus, its limited cpu can't
generate as many pkt as you need (don't know why...)

You should do a test like this one with Iperf:

SOURCEPC-->NODE1-->NODE2-->ENDPC

This kind of test exclude any kind of iperf pkt generation problems.

Bye
Thor

PS:
If you have more questions maybe you can post me directly so I can
reply you in Italian...


On 11/10/06, Roberto Riggio <roberto.riggio at create-net.org> wrote:
> Hi everybody,
>
> We are planning to build a small testbed based on the roofnet code.
>
> In order to do some preliminary testing with the click/roofnet
> architecture we decided to use some already available asus wl500g
> wireless router running OpenWRT.
>
> This router doesn't uses a atheros chipset so I've made a couple of
> changes to the click script in order to make roofnet work on top of a
> device configured in adhoc mode.
>
> Everything is working fine the nodes ping each other and they can reach
> Internet.
>
> However if i try to run an Iperf session between to nodes the click
> process reaches the 99% of the cpu time and the max throughput that I
> can get is 4.6Mbps using TCP over a signle hop.
>
> I've made the same test using some notebook reaching 20Mbps over the
> single hop with the click process at the 20% of cpu time.
>
> Do you have any idea about the reason the performance are so bad with
> the Asus?
>
> Thanks
>
>
>
>
>
> _______________________________________________
> click mailing list
> click at amsterdam.lcs.mit.edu
> https://amsterdam.lcs.mit.edu/mailman/listinfo/click
>