[Click] Counter in userlevel Click

[Beyers Cronje]

Hi Omer,

You should investigate where packets are dropped on m2. At the receive nic,
click queue or tx nic? FromDevice is not ideal, but you should definitely
see better performance than < 50pps, unless you are experiencing live-lock
due to FromDevice still using interrupts.

On 6/23/07, ﻯﻮﻠﻫﺩ ﺮﻤﻋ <omerbeg at gmail.com> wrote:
>
>
> Hi Beyers,
> I have not been able to reproduce the click results. Infact the counter
> reports a very low packet rate. I am running the iprouter.click as a cone,
> the simplest scenario. (A UDP client-server application runs fine between
> m1-m3).
>
>
> m1 (recv)    m2 (iprouter)         m3 (sender)
>  |__________________|  |_______________|
>
>
> --------------- sender.click ----------------
> RatedSource(<...>) -> ToDevice(towards-m2);
> ---------------------------------------------
>
> --------------- sender.click ------------------
> FromDevice(from-m2) -> ctr::Counter -> Discard;
> -----------------------------------------------
>
> And iprouter runs the code which is available from the click site. I
> cannot run PollDevice because I am using D-Link(sundance) cards on all
> machines, and there are no patches for them (hence PollDevice fails).
>
> All machines are running FedoraCore6(kernel 2.6.16). The machines have
> intel celeron 1.7G, with 512MB RAM.
> Is there something I am missing, because the forwarding rates that I am
> getting are < 50packets/s with the iprouter. A simple passthrough
> (FromDevice(eth0) -> Queue(200) -> ToDevice(eth1);) on m2 gives me peak
> rates of ~30,000packets/s.
>
> Thanks in advance for your help.
>
> Sincerely
> -Omer
>
>
>
>
> On 6/17/07, Beyers Cronje <bcronje at gmail.com > wrote:
> >
> > Nope, under linuxmodule you need to access the files under /click or
> > /proc/click
> >
> > I.e. for:
> >  // Click script
> > FromDevice(eth0) -> counter :: Counter -> Discard;
> >
> > Do 'cat /click/counter/rate'
> >
> >
> >