We have SR-IOV based neutron networking for multicast with bare-metal performance. But in this use case, I prefer to use OVS to have multicast traffic isolated within each virtual networks, with overlapping groups, etc.
I am curious how much work it is to write a simple controller to keep track of IGMP join messages sent by mc receivers, and configure a smaller set of destinations on sender VM's OVS instance... On Mon, May 23, 2016 at 11:54 AM, Kevin Benton <blak...@gmail.com> wrote: > From an OpenStack plugins perspective, the Linux Bridge plugin will use a > multicast address address for each VNI for it's broadcast/multicast traffic. > This is still not correct behavior because it will end up going to all > agents in the VNI, but it does result in no fanout from the source node so > it doesn't saturate the source's uplink. > > Unfortunately right now OVS doesn't support targeting a multicast address as > a destination for VXLAN traffic so there will always be a multiplication > factor you have to worry about for broadcast/multicast traffic. > http://openvswitch.org/pipermail/discuss/2016-January/019897.html > > On Mon, May 23, 2016 at 4:13 AM, O'Reilly, Darragh <darragh.orei...@hpe.com> > wrote: >> >> >> > Thanks Darragh. Are you aware of an implementation that treats multicast >> > traffic properly? >> >> I don't, but I didn't research other plugins. >> With ML2/OVS we used VLAN provider networks and let the physical network >> manage multicast between physical nodes. >> _______________________________________________ >> discuss mailing list >> discuss@openvswitch.org >> http://openvswitch.org/mailman/listinfo/discuss > > > > > -- > Kevin Benton _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
