[Click] drops in FromDevice element

[Beyers Cronje]

Hi Guido,

PollDevice does not have a drop counter as packets are queued on the
networks card's rx queue. Use ifconfig to see any drops on the card itself.
If drops occur at the card try increasing RX and TX buffers on both cards
via ethtool.

Beyers

On 11/27/06, Guido Alejandro Gavilanes <ggavilanes at gmail.com> wrote:
>
> Hi Beyers,
> 22322 packets per second. I am testing these conditions and
> performance has been increased comparing with the previous conditions.
> I see that at /proc/click/PollDevice  there is not a drop counter,
> does that mean my card will never drop any Packet?
> Are there drivers for e100 also? I mean, for my other NIC.
>
> Thank you!
> Guido
>
> On 11/27/06, Beyers Cronje <bcronje at gmail.com> wrote:
> > What's your pps rate for PollDevice ?
> >
> >
> > On 11/27/06, Guido Alejandro Gavilanes <ggavilanes at gmail.com> wrote:
> > > Hi Beyers,
> > > I installed e-1000-6.x from cvs, it works, the problem occured with
> > > the other interface who is installed in mi host, which driver is e100,
> > > and polling was not active for that.
> > > thank you!
> > > Guido
> > >
> > > On 11/24/06, Beyers Cronje <bcronje at gmail.com > wrote:
> > > >
> > > > Hi Guido,
> > > >
> > > > The latest polling driver on CVS definately supports Intel PRO
> 1000MT,
> > in
> > > > fact I'm using the same card. You have to compile the latest e1000
> > polling
> > > > driver supplied on CVS.
> > > >
> > > > Typically if FromDevice drops increases it means that the push path
> in
> > your
> > > > click config takes too long before it can process another packet
> from
> > the
> > > > queue, or in the case of FromDevice interrupts are interrupting the
> CPU
> > too
> > > > often which causes delais in the push path processing of click. What
> > > > processor are you using? Increase the ethernet card's RX and TX
> queue
> > via
> > > > ethtool.
> > > >
> > > > But most of all, get your polling driver installed and use
> PollDevice.
> > > > PollDevice is the only polling driver for click and interrupts are
> > > > completely turned off, allowing click scheduling to shine. As
> mentioned
> > > > using FromDevice interrupts are still enabled and there will be a
> > definate
> > > > limit in pps processing due to interrupt locks.
> > > >
> > > > Hope this helps.
> > > >
> > > >
> > > > Beyers
> > > >
> > > >
> > > > On 11/24/06, Guido Alejandro Gavilanes < ggavilanes at gmail.com>
> wrote:
> > > > > Hello Beyers,
> > > > > I use FromDevice, I already tried PollDevice, but it seems my NIC
> > > > > doesnt support polling in the driver, I am using an Intel PRO 1000
> MT
> > > > > Desktop Adapter.
> > > > > Ethernet Frames produced by my tester generator are 64 bytes long,
> in
> > > > > fact verifications of lost frames are being taken from that device
> > > > > counters also.
> > > > > What else it could be?
> > > > > Thank you.
> > > > > Guido
> > > > >
> > > > > On 11/24/06, Beyers Cronje <bcronje at gmail.com> wrote:
> > > > > > Hi Guido,
> > > > > >
> > > > > > FromDevice is not the polling element. Use PollDevice. Also
> 8000pps
> > > > seems
> > > > > > low even for FromDevice, what type of config are you using?
> > > > > >
> > > > > > Beyers
> > > > > >
> > > > > >
> > > > > > On 11/24/06, Guido Alejandro Gavilanes <ggavilanes at gmail.com >
> > wrote:
> > > > > > >
> > > > > > > Hello!
> > > > > > > First I would like to thank you for your support last week. I
> have
> > > > > > > been trying some ethernet tests by forwarding frames from one
> > > > > > > interface to another as before, and I find that with more or
> less
> > 8000
> > > > > > > packets per second on the ethernet input interface, the entire
> > system
> > > > > > > started to loose ethernet frames; So I started to monitor
> input
> > > > > > > ethernet interface and found the drop counter (from
> > > > > > > /proc/click/FromDevice\@7/drops) increasing and
> > > > > > increasing, when I
> > > > > > > slow down the load, slightly, the counter doesnt increase.
> > > > > > > Reading FromDevice documentation I see FromDevice element
> passes
> > > > > > > packets taken from a queue to its output and I was wondering
> if
> > these
> > > > > > > drops are lost due to an overflow in such internal queue.
> > > > > > >
> > > > > > > I tried using BURST parameter in order to have this queue most
> > empty
> > > > > > > as possible on each polling cycle, but I does not seem to have
> an
> > > > > > > effect on observations.
> > > > > > >
> > > > > > > What can I do to decrease these drops? any suggestions?
> > > > > > > Thank you very much!
> > > > > > > Guido
> > > > > > > _______________________________________________
> > > > > > > click mailing list
> > > > > > > click at amsterdam.lcs.mit.edu
> > > > > > >
> > https://amsterdam.lcs.mit.edu/mailman/listinfo/click
> > > > > > >
> > > > > >
> > > > > >
> > > > >
> > > > >
> > > > > --
> > > > > GUIDO ALEJANDRO GAVILANES CASTILLO
> > > > > Electronics and Telecommunication Engineer
> > > > > Universidad del Cauca - Colombia
> > > > > Master in optical Communications and Photonic Technologies
> > > > > Politecnico di Torino - Italy
> > > > > Cell: +39 3405269579
> > > > > Via Sassari 18 Torino, Italy
> > > > >
> > > >
> > > >
> > >
> > >
> > > --
> > > GUIDO ALEJANDRO GAVILANES CASTILLO
> > > Electronics and Telecommunication Engineer
> > > Universidad del Cauca - Colombia
> > > Master in optical Communications and Photonic Technologies
> > > Politecnico di Torino - Italy
> > > Cell: +39 3405269579
> > > Via Sassari 18 Torino, Italy
> > >
> >
> >
>
>
> --
> GUIDO ALEJANDRO GAVILANES CASTILLO
> Electronics and Telecommunication Engineer
> Universidad del Cauca - Colombia
> Master in optical Communications and Photonic Technologies
> Politecnico di Torino - Italy
> Cell: +39 3405269579
> Via Sassari 18 Torino, Italy
>