[ovs-discuss] OVS -- Linux interaction

Abhishek Verma abhishekv.verma at gmail.com
Wed Jul 29 00:44:50 UTC 2015


So youre saying that in DPDK + OVS, OVS actually retains a copy of the
entire routing and ARP tables. Interesting, i hadnt realized that.

About eth0, my assumption was that a port over which a packet needs to be
sent out or received on MUST always be added to the OVS bridge. How can i
trap a packet arriving on eth0, if it has not been added to the OVS bridge?

Thanks, Abhishek

On Wed, Jul 29, 2015 at 6:09 AM, Jesse Gross <jesse at nicira.com> wrote:

> The DPDK path operates differently - the kernel stack in not invoked.
> In that case, eth0 does need to be attached to OVS but, again, not in
> the same bridge. OVS will sync the routing table from the kernel but
> otherwise implements ARP, etc. internally and retains control of the
> packet until it is sent out on the physical interface.
>
> On Tue, Jul 28, 2015 at 5:02 PM, Abhishek Verma
> <abhishekv.verma at gmail.com> wrote:
> > Thanks Jesse.
> >
> > How does this work when OVS is running in the user space along with DPDK.
> > Does OVS again send packets to Linux stack for forwarding? I thought the
> > whole point there was NOT to involve the Linux stack.
> >
> > Asking this since i thought that since eth0 is a part of the OVS bridge,
> its
> > OVS that actually frames the entire packet and sends it out.
> >
> > Thanks, Abhishek
> >
> >>
> >> It does not make sense to have eth0 in the same bridge as the VXLAN
> >> port since they are not in the same network segment (encapsulated vs.
> >> not). OVS simply sends tunneled packets as IP to the Linux stack,
> >> similar to any other application. These packets do not re-enter OVS
> >> unless there is another bridge that is in the normal Linux forwarding
> >> path.
> >
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20150729/455bd8da/attachment-0002.html>


More information about the discuss mailing list