[Click] Polldevice not working with atheros

Adam Greenhalgh a.greenhalgh at cs.ucl.ac.uk
Fri Apr 11 05:58:14 EDT 2008


On 11/04/2008, Giovanni Di Stasi <gdistasi at gmail.com> wrote:
> Il Friday 11 April 2008 02:48:16 Adam Greenhalgh ha scritto:
>
> > To make a device pollable in click requires a fair bit of hacking of
>  > the driver itself, which is probably not worth doing for an atheros
>  > device given its low data rate compared with a 1G wired nic. Why it
>  > kernel panics using FromDevice I'm sorry I haven't got any ideas. Try
>  > user space click with the atheros using from device since I doubt you
>  > need kernel level click for a wireless device.
>  >
>  > Adam
>  >
>
>
> Hi Adam,
>  from my experience you need kernel level click for wireless devices more than
>  for any other device, especially when you cope with embedded systems.
>  With user level click is very difficult to achieve acceptable performances,
>  whereas with kernel level click you have some chance (even though the
>  performances are not very good). The problem is the utilization of the
>  processor that goes up to 100%. Do you think it could be useful to do polling
>  in this circumstances?
>
The processor always goes up to 100% when kernel level click is being
run because the click loop in the kernel is always checking the
desciptors. It does yield though.

Thats interesting, I wouldn't have thought that the wireless device
required so many more cycles. I guess I'm used to running click on
multicore machines where the throughput rates of a wireless device are
in the noise when we are forwarding. Is the wireless device a lot more
expensive to run ? Or is it that the processors are less capable in
this sceanrio.


>  Thank you for your answer... Incidentally, I must thank you and JoonWoo Park
>  for your work on click patch for kernel 2.6.24. I'm using it and it's working
>  quite well.

No problem, glad its being tested.

Adam


More information about the click mailing list