On 2005.11.10 11:47:39 +0100, Grzegorz Nosek wrote:
> 2005/11/9, Herbert Poetzl <email@example.com>:
> > hmm, smells like a mainline issue, to be honest,
> > but if you have time (and the machine) we can do
> > some more detailed investigations ...
> > TIA,
> > Herbert
> I was thinking along similar lines (AMD64 x2 are quite a new brand),
> but I thought that was a hardware issue (slightly malfunctioning CPU
> from an early batch?).
> The box is now probably (I'll know for sure when I get to work) free
> to tinker with so if you have any suggestions on how to debug the
> issue, I'm all ears. The machine is there, so is the time (actually
> this is my priority task at the moment, "stabilise that damn box!") :)
I've tried the instructions you gave in one of your previous mails (add
a qdisc, then add a class [or whatever are the correct terms...]). And
ran "ping -f 192.168.0.1" from my box to my router. With 1Gbit rate and
ceil limits, with 100Kbit limits and backlog filled up and cleaned out
about once per second for a really short time (I guess that is expected
due to the rate limit). I didn't experience the backlog slowly filling
up as you described.
Test setup was:
Athlon64 X2 4400+, Tyan Tomcat K8E S2865G2NR, 2GB RAM, 100Mbit tg3 nic.
Tested kernels (all SMP):
On the vserver-kernels I ran ping in context 0 as well as in a vserver.
Vserver mailing list
Received on Sun Nov 13 17:19:56 2005