On Tue, Oct 19, 2010 at 03:30:15PM +0200, Dirk Bonenkamp - Bean IT wrote:
> Op 19-10-2010 15:19, Graeme Fowler schreef:
> > On Tue, 2010-10-19 at 14:56 +0200, Dirk Bonenkamp - Bean IT wrote:
> >> I still don't know what exactly the problem was with LVS-NAT...
> > Very likely that iptables/netfilter conntrack module was also processing
> > the VIP traffic, adding to the latency. Obviously that's only true if
> > you're using a conntrack module, but this is often the default on NAT
> > directors.
> This wasn't the issue AFAIK. The conntrack modules where not loaded.
> They do get loaded when adding extra NAT rules trough iptables, but when
> only using LVS-NAT, they don't get loaded. The tests where done without
> the modules loaded.
As a heads-up, they will be loaded in 2.6.36, and I expect there
will be some performance penalty as a result. This was an oversight
when the Double-NAT changes were merged. And it will be resolved
in 2.6.37 - the fix has already been merged.
Please read the documentation before posting - it's available at:
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to http://lists.graemef.net/mailman/listinfo/lvs-users