Enterprise multicast design question

[campbech1]

Today we are 3 hospital facilities each utilizing Cisco hardware. The three sites are around 12k devices, 5k devices and 1k devices.

My question is what is the preferred method to setup multicast between sites? Today we have multicast only configured at our main site. I'm looking to enable it at the other sites but I have servers/applications that need to communicate via multicast with clients at all three sites.

My guess is it's preferred to have multiple RPs, one at each site, but then how does multicast routing work between sites? As easy as setting up ip mroute routing entries?

Thank you in advance!

[Peter Paluch]

Hi,

The first question is whether the network interconnecting your sites is itself multicast-enabled. If not, you will probably need to configure tunneling and manually modify the mroute tables to make the RPF check happy, or ask the operator of that network to enable the multicast for you.

The most simple way to start the multicast between the sites once the interconnecting infrastructure is multicast-enabled is to have a single RP, preferably at the main site. Unless you are intentionally using shared trees, Cisco routers will immediately switch over to the source based tree so the RP is unused most of the time.

If you wanted to have multiple RPs, one at each site, then you would need to run the Multicast Source Discovery Protocol (MSDP) between these RPs on each site, similar to an application called Anycast-RP:

http://www.cisco.com/c/en/us/td/docs/ios/solutions_docs/ip_multicast/White_papers/anycast.html

Setting up ip mroute routing entries would absolutely not help here because what you need is having one RP inform other RPs about existing sources to given multicast groups so that the other RPs can build source trees to the discovered sources. MSDP is easy to set up with two RPs; using more RPs may need certain tweaking as MSDP is closely tied to BGP (it was originally invented for inter-AS multicast).

Yet another possible solution would be to run Bidir-PIM. In Bidir-PIM, the RP functionality is extremely reduced - the IP address of the RP is merely a direction pointing "upstream" in the distribution tree but there is no multicast registering and tunneling involved. It can safely be any address in a network advertised in a routing protocol but that IP address does not have to be assigned to any node.

Feel welcome to ask further!

Best regards,
Peter

[campbech1]

Peter,

Thank you very much for your reply.

The links between the sites are multicast capable and if the design is acceptable I'd like to stick to a simplistic design of a single RP.

So by the sounds of it, since we have a RP defined at our main site, just need to enable PIM-SM to/from the main site core switches to the WAN edge routers and from the WAN routers to/from the other sites core switches?

[Peter Paluch]

Hello,

You are welcome.

Regarding the design with a single RP - I suggest starting with it, and should it ever prove to be unsatisfactory, we can always add new RPs and mesh them with MSDP. However, starting with a single RP is the easiest way to try to get this running.

So by the sounds of it, since we have a RP defined at our main site, just need to enable PIM-SM to/from the main site core switches to the WAN edge routers and from the WAN routers to/from the other sites core switches?

Yes. Basically, each router through which the multicast will flow from the sender to the recipient must have ip multicast-routing or ip multicast-routing distributed configured, and all its interfaces that will carry the multicast must have ip pim sparse-mode configured. In addition, all these routers must be equipped with the IP address of the RP - assuming you are not using Auto-RP or BSR to dynamically advertise the RP's address, use the ip pim rp-address address command to statically configure the RP's address on each router.

Best regards,
Peter

[campbech1]

Peter,

Can't thank you enough. I have a change window tomorrow that I'll implement that design and start testing between the sites.

I'll post tomorrow what the outcome of implementing this is.

[Peter Paluch]

Hi,

Wonderful - please let me know how all of this worked out for you tomorrow.

Best regards,
Peter