mGRE over MPLS cloud for multicast

[gnijs]

Hi all,

Our companies backbone runs over an international MPLS cloud. Every site runs locally EIGRP and the cloud itself runs BGP. All ok, however this MPLS cloud is not multicast-enabled.

I was playing with the idea of implementing dynamic mGRE tunnels (ipsec is not necessary since MPLS is private VPN) to carry the multicast traffic only ON TOP of the normal IP traffic in the MPLS cloud. Multicast should follow the tunnels, normal traffic should use the normal links. I need some sort of dynamic build up of tunnels when multicast needs to be delivered to another site. Any site can be a MC receiver or sender, so full mesh capability is required.

Is this possible and are there any references to a scenario like this ?

regards,

Geert

[mheusinger]

Hello Geert,

there is one issue I would assume: Reverse Path check (is multicast stream received through interface with best metric to reach unicast source/server address) will fail in case unicast traffic for your servers will not be routed through the tunnels.

So I would recommend to use the MPLS VPN multicast solution offered by Cisco to avoid this problem.

Hope this helps! Please rate all posts.

Regards, Martin

[ekiriakos]

Hi Martin,

I may be wrong with this one, but I think if the GRE tunnel is made passive in the IGP process and multicast static routes point to the tunnels , then unicast traffic will flow through the physical links and m/cast over the tunnels. I believe this should work, at least for static tunells anyway...

Rgds

E.

[gnijs]

Hi Martin,

First, the MPLS is provided by a SP. We do not have the ability to implement new things in the cloud, however, we do manage the CE routers.

RP check is also what i was wurried about. The physical MPLS interface normally has the best metric. Only for multicast, the tunnel should be taken. This can be achieved by not running PIM on the physical MPLS links but only on the tunnel interface.

On each side you will need:

ip mroute 0.0.0.0 0.0.0.0 Tunnel0

to specify that multicast is to be expected from Tunnel0 to pass the RPF check.

Geert