[ovs-discuss] OVS bridge implementation considerations for physical switch

Joe Beck jbeck at brightwaysco.com
Thu Jul 11 00:21:16 UTC 2019


I’m just beginning to experiment with different configuration options and am looking for any docs or tips people may have relative to our application of OVS. Since we have our own switch silicon we obviously don’t need DPDK. At this point, we’re strictly looking to leverage OVS to manage the control plane.
Still not deep enough yet to know whether we should be more focused on ovsdb protocol rather than OVS so need to continue to educate and experiment.
Part of the challenge has been wrapping my head around our use-case vs the examples & docs out there. For example, since its all just control plane I’m assuming all ports should be set to type=internal. And we should set all these physical network ports as part of 1 large bridge as I think there may be advantages to that? Such as when it comes to VLAN options & building that interface into our physical switch.

The plan is to continue to add L3 and other capabilities to our switch so leaning more toward OVS at this stage.

More context—
We’re a few years into the development of a physical switch. The initial release is focused on L2–10/40/100G.
We’ve written custom C pgms for MAC Learning/propagation with the intention to integrate with OVS to makeup our control plane/NOS.
It’s a p-spine fabric connecting up to 16 line cards each comprising up to 52 ports. With such a large # of interfaces, I’ve been researching the best approach for our OVS implementation. For this hardware switch scenario, I haven’t found much material out there. One thing I recently found is:
http://docs.openvswitch.org/en/latest/topics/integration/
So again, we’re early in our research & will be trying various configurations and options but I’m hoping others may have some hints or suggestions.
Regards,
Joe
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openvswitch.org/pipermail/ovs-discuss/attachments/20190711/35daa0e0/attachment.html>


More information about the discuss mailing list