Re: netfilter queue throughput slowdown

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi,

I upgraded the kernel and distro but some only a little improvent. I
saw no failures to set the verdict, but many many packets drops in the
netfilter queue proc entry.

I saw the two submitted patches today, and will try those.

I think having an option on what to do with stuck packets makes a lot
of sense. I can think of several ways to handle this, and each have
different consequences depending on the type of packets in the queue.
Having them stuck in the buffer is really bad though.
If you do a massive ACCEPT, packets will probably arrive out of order,
which may cause problems for some protocols and apps. If you do a
massive DROP, TCP will deflate its window and TCP will cut down its
rate significantly. Here it might be better to some RED type dropping,
maybe successive single DROPs untill queue is empty of stuck packets.

Thanks,
Anders

On Thu, Jun 30, 2011 at 7:45 PM, Eric Dumazet <eric.dumazet@xxxxxxxxx> wrote:
>
> Le jeudi 30 juin 2011 à 19:07 +0200, Eric Leblond a écrit :
>
> > As the verdict failure is bound to occur in a high load time,
> > retransmission of the verdict (which is necessary) will not help the
> > system to recover. Userspace has to deal with it but it has another
> > consequences which is that userspace software may suffer of case where
> > successive failures occurs.
> >
> > In this scope, Florian's patch "netfilter: nfqueue: batch verdict
> > support" could be really useful. It could be used by userspace to
> > trigger an decide on all stucked packets. Issuing a massive ACCEPT could
> > lead to dynosaurus packet coming from ancient time but it could be ok if
> > batch occurs enough often.
> >
> > Is there a plan to accept it in mainstream ?
>
> Given that apparently some apps are not aware some of their verdicts are
> lost, I consider the BATCH idea would be a bad idea, unless DROP is
> used.
>
> If you have any doubt, only sane thing is to drop packets, not accept
> them.
>
> Maybe a single queue flag is needed : DROP_OLD_PACKETS, if user
> application is handling packets in order.
>
> Every time a verdict is given by application, automatically DROP all
> previous un-verdicted packets.
>
>
>
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[Index of Archives]     [Netfitler Users]     [LARTC]     [Bugtraq]     [Yosemite Forum]

  Powered by Linux