[PATCH] Re: Oops in 2.3.99pre9-2/SMP
Fri, 2 Jun 2000 16:26:17 +0200 (CEST)
On Fri, 2 Jun 2000, Rusty Russell wrote:
> OK, reproduced this, and after much profiling, I traced it down to
> massive hash chain collision in the NAT code (well, other than a
> debugging message about UDP checksums which will go away for 2.4.0).
> It turns out that we don't need to traverse the chain anyway for most
> cases: the patch in CVS fixes this, and it's much nicer. Note that
> with CONFIG_NETFILTER_DEBUG on, it's still a significant slowdown, but
> without that it's fine under 100baseT and 10baseT floods.
> See CVS for patch against 2.4.0-test1.
I'm testing the patch right now and everything is very happy, especially
me :) I'm running udpsic, tcpsicp and icmp, all att the same time here and
everything just works as I want it to.
Rusty, any chanse that we can increase the UDP timeout from 60*HZ to
120*HZ or maybe even 180*HZ ?
My lixq timesout with 60, I'm running with 180 right now and everything