On Wed, Oct 12, 2005 at 02:39:07PM +0200, Luca Maranzano wrote:
> Hi Christophe,
>
> the long thread was started by me. I still don't have the hardware to
> test the patch, even if I can report that the current system is
> working fine despite the load average is 2.0.
>
> It seems that I'm the only one on this list using the LVS whith
> ipvs_syncmaster and ipvs_syncbackup to have failover and failback in
> an HA production environment ;-)
>
> Are you experiencing a similar problem? Which is your setup?
>
> If you read the post you will note that the problem arises apparently
> only on Xeon CPU.
I do have some additional information, and probably good news.
When I originally tested this problem on my non-Xeon P4, I was
looking for 100% CPU utilisation, which I do not get.
However, subsequent discussion has pointed out that it
is actually a load and not a CPU problem. I do indeed get a load
of around 2 when I run both daemons.
This is good, because we know know that the problem is not
some weird Xeon problem, that frankly made little sense.
Its just a timer issue, and we just need work out
out which of the two patches I posted earlier is best.
To produce the problem all you need to do is run:
uptime
sudo ipvsadm --stop-daemon master
sudo ipvsadm --stop-daemon backup
sleep 300
uptime
sudo ipvsadm --start-daemon master
sudo ipvsadm --start-daemon backup
sleep 300
uptime
Well, of course you need to recomile the kernel and reboot.
I'm pretty busy right now, so it would be a tremendous help
if someone could test the patches. I can work on getting
one of them included there.
Ohh, and the 300 seconds is a bit arbitary, its just
time for the load to cool down and warm up.
--
Horms
|