[c-nsp] Internet in VRF

[Mark Tinka]

On 4/May/15 04:52, Nathan Ward wrote:
>
> These two statements seem to conflict, though I may be missing context
> or reading them wrong. You use MPLS for TE, but do all your signalling
> in LDP and not RSVP? Can you expand on that? Do you do CR-LDP or
> something?

We do use RSVP for TE, but only in 0.001% of cases. Such use-cases
require explicit paths to be set around the network, and given they are
very, very corner cases, I can live with the RSVP pain. This is where,
for me, MPLS is reasonably useful, in addition the NG-MVPN and l2vpn's.

>
> Worth noting, TE does not require that you run MPLS on top. You can
> quite happily put IP/IPv6 packets in to TE tunnels. I am a big fan of
> Internet in a VRF/L3VPN, but, if you want to run TE without L3VPN you
> certainly can do so.

We don't run l3vpn's over RSVP. LDP handles that.

I have one very corner case where we run an l2vpn inside a TE tunnel,
but that falls in the 0.001% bracket, so is acceptable.

Everything else follows the LDP path.

>
> Just the same, you can stick LDP labels on IP/IPv6 packets and have
> them forwarded, without any VPN labels on top.

So VPN services are where we sell l2vpn's and l3vpn's.

But all global IPv4 traffic is MPLS switched, because we run a BGP-free
(IPv4) core.

IPv6 traffic is natively forwarded using regular hop-by-hop
routing/forwarding (I don't believe in 6PE), and once LDPv6 gets support
on all our platforms, I'll take BGPv6 out of the core and have that
traffic MPLS switched also.

And then I'll wait for SR to get usable enough for me.

Mark.