On Mon, 2005-10-24 at 19:29 -0400, Chuck wrote:
> > > ok.. so how may ips in a guest do you estimate could be used before a
> > > perceptable change in performance happens? and by this are we talking
> about
> > > per interface or gross totals among all interfaces?
> > That really is a hard question to answer, and a research topic. "Suck
> > it and see", or build yourself a test system so you can definitively
> > answer the question.
> > It's quite possible that on today's typically CPU-heavy systems it might
> > not be a perceivable bottleneck at all until you start getting to the
> > kind of connection levels where NAT starts not being able to track
> > connections any more. If you can measure any performance difference as
> > the number of IPs in a vserver goes up at all I'd be highly impressed.
> ok I was just being concerned that if I set up a guest with 10 or 15 ip addys,
> as each gets heavily used by virtual sites, the virtual machine may take more
> time than it would under a stand-alone hardware situation.
Apples and Oranges.
A more relevant question is, if you had a single machine without
vserver, and with several Apache instances, each set to listen on 10 to
15 addresses - is that faster or the same speed as a system using
vserver, with each block of 10 to 15 addresses assigned to a vserver,
and apache running inside that vserver. Assuming you're using
filesystem unification, there is virtually nothing different between
those two setups - just the vserver overhead.
> we dont use nat,
> every ip is public.
yes, sure. It was just drawing attention to the fact that TCP itself
has performance limits, and that the current implementation of virtual
IPs might possibly be fast enough that you'll run into those other
limits first.
Sam.
_______________________________________________
Vserver mailing list
Vserver@list.linux-vserver.org
http://list.linux-vserver.org/mailman/listinfo/vserver
Received on Tue Oct 25 04:16:40 2005