[ovs-discuss] Openswitch VM to VM traffic switching
blp at nicira.com
Wed May 19 04:02:09 UTC 2010
On Tue, May 18, 2010 at 10:39:15PM -0400, Todd Deshane wrote:
> Maybe it would be useful to have a INSTALL.KVM?
> It would be the same as INSTALL.Linux, except for at the end, a
> description of using the custom ifup and ifdown scripts on the kvm
> command line.
I agree that adding an INSTALL.KVM is a good idea. I think that we at
Nicira have limited experience with KVM integration yet. Is there any
chance that you could start us off with a draft for this new file? I'll
gladly add it, if I have something to add.
Instead of duplicating (most of?) INSTALL.Linux, I would rather refer to
it. Then, if it changes, it only has to be changed in one place.
> Also, I am in the middle of doing some benchmarking for my research. I
> did a quick comparison of standard bridging to open vswitch using
> iperf that showed no degradation from open vswitch in the tests I have
> ran so far (bridge in base to remote linux, KVM guest to remote linux,
> and KVM guest to base bridge). I am using Open vSwitch version 1.0.0
> in my tests.
That's good to know.
There are some benchmarks, of an older version of OVS, in the paper
posted on the Documentation page at openvswitch.org.
It's been a while since we've carefully looked at OVS numbers versus the
Linux bridge. If you find any big discrepancies, please let us know, so
that we can see about fixing them.
(Have you tried making a graph like that in the OVS paper? I am curious
to see whether others also see the anomalies shown in the graph at 192
kB and 4 MB transfer sizes.)
More information about the discuss