BGP and Multicast

[fsebera]

We have an MPLS network. MPLS is handled by Telecom vendor.

Each of our spoke sites peers with a single Telecom BGP peer (CE-PE).

We would like to enable Multicast on our network. We thought using Multiprotocol BGP Extensions for IP Multicast which seems would allow us to bypass the Telecom MPLS provider. We don't want to involve the Telecom with this configuration.

Looking for advice.

Tks

Frank

[Peter Paluch]

Hello Frank,

Running MP-BGP alone will not make the multicast possible. What you need is a multicast-capable transport between your spokes sites. If you do not want to ask Telecom to provide you with a Multicast VPN service then you will most probably need to resort to tunneling the multicast. However, if the number of spoke sites is large, this may not suit your needs. How many spoke sites do you need to interconnect using multicast and how strongly does their current count changes over time?

Best regards,

Peter

[fsebera]

Ok, additional research has been completed.

Our telecom provider (lets call the them -V) is baulking at the idea of enabling multicast for us. We are being told by their telecom engineers that their PE and P routers are not Cisco (if anyone can believe that). While at the same time, we are looking to replace some of our older Cisco routers. Yes we had to consider other vendors (lets call the other vendor -J) but as it turns out, the other vendor does not support Bi-Dir PIM / Phantom-RP which is what we want/need to run. Now the telecom indicates that also have this vendor's equipment as their PE and P infrastructure.

What do you think?

[Giuseppe Larosa]

Hello Frank,

if you can move to PIM SSM mode the J vendor routers support NG MVPN very well, I can say this because I took part in a project and we developed an european wide SP network and we used these J- routers with NG MVPN to support high volume video streams distribution.

Clearly, SP V- should accept to configure its network to support NG MVPN  (it requires a special address family inet-mvpn in MP BGP configuration configured on PE nodes and route reflector servers)

The use of PIM SM would require some additional hardware on some J routers (on RP and on PIM source DR a tunnel PIC would be needed)  so PIM SSM has to be preferred.

Hope to help

Giuseppe

[Mohamed Sobair]

Adding to Peter's wonderful reply,  is that using GRE Tunneling to transport Multicast traffic for End Customers varies. what I mean here is that this solution have drawbacks , one of the main drawbacks is bandwidth limitation.

So, if you dont need to envolve your SP in such situation, make sure you dont have high bandwidth requirement.

Regards,

Mohamed

[fsebera]

Thanks guys!

We have 300 sites.

Multicast traffic is low-end at about 1MB max at any time.

It appears Cisco docs say this is how we could set this up.

CURRENT SETUP

CE-----eBGP----PE_WAN_PE-----eBGP-----CE

NEW SETUP added to the current setup above.

CE------------------------eBGP------------------------CE

Each of our sites is configured in a different AS while the cloud is a single AS.

Thanks for the assistance

Frank

[Peter Paluch]

Frank,

I believe that the primary misunderstanding is that you think the BGP is going to carry your multicast traffic. With absolute certainty, it is not. In fact, you do not need BGP at all to have multicast; the BGP is used mostly when doing interdomain multicast or when different unicast and multicast topologies are required. The main multicast routing protocol you will need is PIM.

However, the main point stands. Regardles of routing protocols in place, your multicast must so or so traverse through the Telecom network. Now, if the Telecom network does not know where to send and replicate the multicast, you won't be able to pass multicast through to your spokes. Your only two options are tunneling, which is in my opinion unrealistic with 300 spoke sites (imagine a router replicating each multicast packet to 300 spoke sites!), or actually asking Telecom to provide multicast VPN service for you.

Best regards,

Peter

[Giuseppe Larosa]

Hello Frank,

with 300 sites any solution that involves packet replication over GRE tunneling is not feasible: a 1 Mbps multicast stream would mean 300 Mbps of traffic exiting the HQ and directed to the spokes!

As Peter has explained MP BGP does not solve the problem of how to forward multicast traffic between sites, it may be of help in the signalling plane, but it does not provide alone any transport mechanism.

The best move you can do is to contact your ISP and ask them to provide you an offer for a multicast VPN service for all 300 sites.

SP provided multicast VPN provides customer multicast traffic replication in the SP backbone adding PIM at PE-CE links. You can keep the RP(s) in your routers or have RP role played by some PE nodes.

No additional or dedicated PE-CE links are needed unless the provider wants to use a VPLS service for the multicast traffic replication.

To be noted in addition to draft Rosen multicast VPN that uses multicast GRE in SP networks to transport customer multicast, NG MVPN has been deployed in the last years and it uses P2MP MPLS LSPs.

Hope to help

Giuseppe

[fsebera]

Never expected BGP to carry multicast traffic. We see the MBGP Extensions for IP Multicast it is possible to configure BGP peers to exchange unicast and multicast NLRI which will allow PIM to ride the BGP extensions. Or we could configure GRE and/or Multipoint GRE tunnels to also carry the multicast PIM traffic. Either way, we are looking for a way (if the telecom and our corp cannot reach an agreement) to bypass the Telecom in our setups. A 300-site setup is not excessive by any means. I worked for an International corp in the past with several GRE tunnel links in every country around the world, each link carring multicast traffic and that worked just fine. There is planning and maintance required as expected but not a complete burden.

Since Cisco is so flexible, I know there are probably more solutions to this issue than the two listed above. AND this is what we were aimming to resolve.

Thanks for offering a different look!!

Frank

[Peter Paluch]

Hi Frank,

We see the MBGP Extensions for IP Multicast it is possible to configure  BGP peers to exchange unicast and multicast NLRI which will allow PIM to  ride the BGP extensions

That is true but why do you focus on BGP so much? PIM is "Protocol-Independent Multicast", meaning that it is independent of what protocol built the routing table - and it truly is. From what you have indicated so far, the BGP would not carry any information that is not already available in your current service, either by static or unicast dynamic routing. Do I miss anything here?

A 300-site setup is not excessive by any means.

For unicast, it is not indeed. But please take into account that if the Telecom is not going to provide multicast-enabled service, all GRE tunnels, even if multipoint, will always be unicast. That means that the egress GRE tunnel device will have to replicate the egress multicast itself and send it in unicast-addressed GRE packets. A single 1Mbps multicast stream will need to be replicated into 300Mbps worth of GRE traffic. Do your devices have such power and fat uplinks to the Telecom to cope with this burden?

Since Cisco is so flexible, I know there are probably more solutions to this issue than the two listed above.

There may be other technologies I am not aware of, that is for sure. But irrespective of the particular technology, please try to see the technology from the pure theoretical viewpoint: if the transport network is not multicast-capable, you will have to reconvert multicasts into unicast to be able to carry them through the transport network. As soon as you do that, you end up with creating a NBMA overlay network with the replication burden moving to a single or a small number of devices (just thinking of BUS server from ATM LANE). There is simply no escape from that - correct me if I am wrong here but this is what the fundamental principle says.That poses, in my opinion, strong limitations to scalability.

Best regards,

Peter

[Mohamed Sobair]

Frank,

As already denoted, the P-t-P GRE Tunnels have drawbacks , especialy in your case is the headquarter side.

Peter made a nice explanation, I would also add another point is that, there is no way to have Multipoint GRE Tunnel as a Solution.  The Multipint GRE tunnel is a lone cant be implemented because there is no a tunnel destination as you know.  Its Cisco Specific that they introduced DMVPN, which uses NHRP protocol for Spokes to register with the Hub over an IPSec Tunnel knowing that Encryption adds another layer of Security and of course performance decrease to the normal traffic.  For 300 Sites, its not practical.

This is Why, Service Providers have Multicast VPN (MVPN) Solution for such requirement. I would highly advice of running your multicast through the MPLS Provider by getting them involved. Keep in mind the Service Provider itself is independant for your Internal Multicast Traffic, it primarily use its infrastructure to transport your Multicast, You can Keep your RPs and control the Multicast traffic within your internal Network. This is a feasible and a practical solution for your requirement. UNLESS you dont want to pay money and want to establish 300 GRE Tunnels which I doubt your Head quarter router can handle along with the normal traffic that its handling. If you have really a Capable hardware at the HQ, then go for it.

Regards,

Mohamed