It would be nice to be able to specify the L2MTU and encapsulation type for BGP signaled VPLS, just as you already can for LDP signaled VPLS. The ability to filter L2VPN routes would also be nice, as would the ability to examine advertised/received L2VPN routes.

Finally, and most importantly to me at the moment, a mechanism for verifying individual VPLS tunnel functionality before making it active would be great. Perhaps BFD could fill this role.

This feature would be especially useful for multi-homed VPLS where (R)STP is unusable. When there are multiple routers in a VPLS at a single location, which use the same RD and "Site ID", the generated NLRI for that VPLS will be the same. Therefore BGP route selection will suppress intra-location tunnels, and will only build a single tunnel to each remote location.

This is great, except that every router at a given location will build that single tunnel. Only one will be bi-directionally functional, but at the moment ROS tries to make use of all of them, leading to half-loops (brodacast/multicast packets being returned to their sender intact, confusing bridge learn tables in the process). If it were possible to suppress tunnels that aren't bi-directionally functional, then a loop-free topology could be created without using (R)STP.

This is actually the suggested form of loop-control for multi-homed VPLS in section 3.5 of RFC-4761 (along with STP), and also appears in Juniper documentation.

Thanks,
--Eric

I just noticed that 'pw-mtu' was added to the BGP-VPLS configuration.

Thanks guys!
--Eric