[c-nsp] m-vpn

[Mark Tinka]

On Monday, July 02, 2012 09:28:15 AM adam vitkovsky wrote:

> Right the need to peer with every other PE in the mVPM
> domain -but what would be the limitation on the PIM
> adjacencies on modern PE boxes 1k maybe 2k?

Well, obviously with increase in control and data plane 
resources, you can scale better, but it doesn't necessarily 
mean it's still a good solution.

The reason I like NG-MVPN (or BGP-MVPN if you work at 
Juniper) is that, if you think about it carefully, it's a 
Unicast operation in the core. That lends itself well to the 
l3vpn concept, which is why it scales very well, decoupling 
control plane functions from data plane functions, unlike 
classic Rosen MVPN's.

p2mp RSVP-TE and mLDP actually create Unicast LSP's, and the 
NG-MVPN infrastructure simply replicates traffic across 
those Unicast LSP's in a way that looks like Multicast (so-
called branch routers, mid-point routers and bud routers). 
It has good scaling properties, especially if you're running 
mLDP.

On Cisco platform like the CRS, even though the MPLS data 
plane signaling is Unicast in nature, it still takes 
advantage of the fabric replication architecture within the 
chassis itself, which is good.

> Yes that would mostly apply in mVPN implementations for
> VPLS, however you'd need PIM at both ends for L3 VPNs
> where you are basically extending customer's m-cast
> domain over the your MPLS core

You only really need PIM on the Sender PE-to-Sender CE link.

All customers connected to Receiver PE routers only need to 
support IGMP. But like I said, many operators run PIM on the 
Receiver PE routers anyway as well, since it enables IGMP 
automatically and simplifies global device configuration. 

BGP takes care of distribution of PIM data across the 
backbone, negating the need for PIM in the core.

MPLS takes care of forwarding of Multicast traffic from 
Source to Receiver, negating the need for IP/GRE in the 
core.

> Hahaha right with my comment I haven't meant it quite
> like that :)

My point was that BGP is being asked to do a lot. Some of it 
makes sense, some of it doesn't. I'm not deluded, but I 
appreciate a good solution when I see one :-).

> That's great news, but still it would be great to see the
> support for the 7200 -but I guess it's not going to
> happen as they would like to force us to upgrade to
> ASR1k

You won't see it for the same reasons you didn't get VPLS on 
the 7200 platform. The most you'd ever see on the 7200 
platform re: NG-MVPN would be MCAST-NLRI support if it's 
being used a route reflector (which is what happened also 
with the VPLS l2vpn NLRI on the same platform), but that's 
all you'll see.

You'll need to invest in the ASR1000 or ASR9000 if you want 
full NG-MVPN capability, from Cisco.

Hope this helps.

Mark.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
URL: <https://puck.nether.net/pipermail/cisco-nsp/attachments/20120702/6643fc81/attachment.sig>