Re: why do we add idle->Todevice with each PollDevice ?

Could you please explain what was the reason of using Idle -> ToDevice in
the past ?
 I am trying to measure the CPU cycles consumed by PollDevice and it seems
like  Idle -> ToDevice has an effect on that, i.e.  Having   Idle
-> ToDevice increases the speed of Polling (number of polls per second)
when you have no traffic, so you will have more empty polls, but as soon as
you remove Idle -> ToDevice  the number of empty polls decreases, I am
still talking about no traffic case. When you have traffic, Idle
-> ToDevice has no effect at all and the Packets/sec is the same either you
have it or you don't, also the number of empty polls doesn't change.

By the way I am using e1000 click patched driver.

Ahmed

On Mon, Nov 14, 2011 at 5:28 PM, Eddie Kohler <[email protected]> wrote:

> (1) If you add "Idle -> ToDevice(X)", that shouldn't actually take
> very much CPU.
>
> (2) BUT you are right. I think maybe this message is left over from an
> older version of Click. Is anyone using polling device drivers? Can
> anyone confirm whether or not ToDevice helps performance in a
> sniffer-only PollDevice configuration?
>
> Eddie
>
>
>
> On Mon, Nov 14, 2011 at 9:24 AM, ahmed A. <[email protected]> wrote:
> > hello,
> >
> > Please could any one explain to me why do we add Idle-> ToDevice (eth0)
> to
> > a configuration file, when we have PollDevice(eth0) running on the same
> > interface ?
> > I got warning messages saying without the Idle-> ToDevice (eth0), I will
> > get Bad performance , why is that? I would expect adding a ToDevice will
> > consume some
> > CPU cycles and degrade the performance rather than improving it ....
> >
> >
> > I would appreciate any help or tips.
> >
> > Thanks,
> > Ahmed
> > _______________________________________________
> > click mailing list
> > [email protected]
> > https://amsterdam.lcs.mit.edu/mailman/listinfo/click
> >
>