[c-nsp] Number of route reflectors, best practice?

[Mark Tinka]

On Sunday, July 24, 2011 10:30:24 PM Gert Doering wrote:

> What's that?  Genuinely curious - haven't done anything
> with multicast in a while, and nothing with
> multicast+MPLS ever.

NG-MVPN (Next Generation Multicast VPN).

It's different from the commonly-known/used Draft Rosen MVPN 
in that NG-MVPN is LSM-based (Label Switched Multicast).

In Rosen MVPN's, PIM and IP/GRE are used everywhere in the 
network (including the core). NG-MVPN is different. It 
utilizes MPLS in the core to replace IP/GRE, and uses BGP at 
the edge to replace PIM (funky, right?).

The MCAST-VPN NLRI is used in NG-MVPN's to perform the exact 
same task that PIM would. In NG-MVPN networks, there is no 
PIM in the core, so BGP is used to transact the PIM control 
plane between edge routers throughout the network, i.e., 
when a downstream user requests a channel change via IGMP, 
that request is sent toward the source via BGP - a process 
normally handled by PIM.

The forwarding plane is signaled by RSVP utilizing p2mp 
LSP's (Cisco and some others are already adding support for 
mLDP - Multicast LDP - supporting both p2mp and m2mp). Once 
RSVP signals the p2mp LSP between the Sender PE router 
(router connected toward the Multicast Source) and Receiver 
PE router (router connected toward Multicast end-users), 
MPLS forwards all Multicast packets and replicates them at 
the network level using the p2mp construct. Cisco's newer 
platforms take this a step further and actually replicate 
p2mp-signaled traffic in the fabric as well.

Unfortunately, AFAIK, only Juniper support the NG-MVPN 
infrastructure, even though many vendors today do support 
p2mp RSVP LSP's. We've successfully inter-op'ed a Cisco IOS 
XR core with a Juniper edge for NG-MVPN; but then again the 
core only needs to be p2mp-aware, i.e., the core isn't VPN-
aware.

NG-MVPN gives us much more freedom in quickly deploying 
Multicast networks because we already have MPLS running. So 
we don't need to enable PIM or GRE in the core, nor do we 
need special Juniper Tunnel PIC's for PIM (although I could 
tell you unending horror stories about this, hehe).

But it does mean that for the foreseeable future, only 
Juniper may support the VPN portion of LSM. I'm pushing our 
Cisco account team to bring it to the ASR9000, and just like 
Cisco folded and had to end up supporting BGP-based VPLS 
signaling in the ASR9000 to compete with Juniper's MX line, 
I don't doubt they'll provide full NG-MVPN support soon. I 
also don't doubt that Juniper will support mLDP in the 
future. This is yet another Cisco-Juniper battle of BGP vs. 
LDP vs. IP vs. MPLS.

Hope this helps.

Cheers,

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/20110724/bbd47fdc/attachment.pgp>