Contrail implements a multicast replication tree based on top of the unicast-only underlay. In 1.10, the limitation is that you can only have a pair of control-nodes signaling these trees. Recent versions of the software are capable of stitching together trees managed by different control nodes.
By default we allocate a single multicast group per virtual-network, used by the link local multicast group (224.0.0.0/24) and by IP broadcast (255.255.255.255 and subnet directed broadcast). There is no mechanism today that allows you to define a multicast tree. Potentially we do implement an IGMP/MLD querier or allow the user to configure multicast group membership on neutron ports the way that IP unicast addresses are configured. Most people I talk to are in favor of the later model. People are using multicast today inside the overlay in order to enable some applications that rely on IP multicast for discovery (elastic search tends to come up as the typical example). You can typically configure the multicast group that these applications use. There are people than do tend to ask about having a multicast source inside the overlay distributing traffic to receivers outside the overlay. One possibility is to give these sources an L2-only network (so that the IP semantics are not interpreted) that then has a leg on an EVPN capable box (e.g. Juniper MX). This DC gateway can then be the multicast designated forwarder for the EVPN lan and originate the multicast traffic. I’m not aware of anyone that has this scenario deployed in production (elastic search in is prod in a couple of places) but that may be just that i don’t know about it… in theory "it should work”(tm). > On Jun 18, 2015, at 7:19 AM, Foucault de Bonneval <[email protected]> > wrote: > > Hi all, > > What are the contraints for using multicast inside the overlay ? > > When I "ping 224.0.0.0" from a VM, I get the packet on the other VMs in the > same subnet. > > I can go up to 224.0.0.255, but 224.0.1.0 does not propagate to the other > instances of the same network. > > Is it a 1.10 issue or maybe not an issue ? > > Thanks; > Foucault > _______________________________________________ > Dev mailing list > [email protected] > http://lists.opencontrail.org/mailman/listinfo/dev_lists.opencontrail.org _______________________________________________ Dev mailing list [email protected] http://lists.opencontrail.org/mailman/listinfo/dev_lists.opencontrail.org
