[ovs-discuss] Even with ovs-controller vswitchd is still heavily used

[kevin parker]

Mostly HTTP,

For confirming bench marked Apache with ab, ovs-vswitchd reached 100% and
ovs-controller 14%

ovs-dpctl show at the time of benchmark

system at xenbr0:
lookups: hit:55391 missed:966296 lost:0
flows: 0
port 0: xenbr0 (internal)
port 3: vif2.1
port 4: vif1.1

Do i need to make any changes in ovs-controller.

Thanks,
Kevin.


On Wed, Jun 5, 2013 at 9:56 PM, Ben Pfaff <blp at nicira.com> wrote:

> On Wed, Jun 05, 2013 at 07:11:44PM +0530, kevin parker wrote:
> > Hi,
> >
> >      i have  ovs-controller with default setting  (ovs-controller -v
> > ptcp:6633).
> >
> >
> > on xenserver host with ovs
> >
> > *ovs-vsctl get-fail-mode xenbr0*
> > *secure*
> >
> > I set it to secure mode to test that host should not use vswitchd for
> flows
> > and forward it to controller for generating flow .*
> > *
> >
> > *ovs-vsctl get-controller xenbr0*
> > *tcp:10.10.10.183:6633*
> > *
> > *
> > * ovs-dpctl show
> > *
> > *
> > hit:8560336 missed:1060122 lost:0
> > flows: 81
> > *
> >
> > where xenbr0 is an isolated bridge with no connection to host interfaces.
> > No other bridge is connected to this controller.
> > When i ping vms on same  host l can see  ovs-vswitchd using 35% CPU.
> >
> > and ovs-controller almost 11% CPU and some times non responsive ssh
> session
> > on hosts.
>
> What kind of trafffic do you have in your network?
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20130606/27a0ab90/attachment.html>