[c-nsp] tag-switching advertise-tags for?

Oliver Boehmer (oboehmer) oboehmer at cisco.com
Tue Oct 25 02:59:30 EDT 2005


Jon Lewis <> wrote on Tuesday, October 25, 2005 12:37 AM:

> On Mon, 24 Oct 2005, Robert Kiessling wrote:
> 
>>> tables or policy routing.  Do all the PE router lo0's have to be
>>> advertised into TDP/LDP for MPLS-VPN tag switching to work?
>> 
>> All your loopbacks that are nexthops for BGP in the vpnv4 address
>> family have to be advertised by LDP, and only those have to.
>> 
>>> How/why would traffic to bgp learned non vpnv4 prefixes be
>>> label-switched if tags for those prefixes were not advertised?
>> 
>> If the next hop is label switched, then the prefix itself
>> is too.
> 
> That sucks...so getting our non-MPLS traffic routed instead of
> tag-switched is actually going to be much more complicated (or at
> least more work) than just tag-switching advertise-tags for <acl> on
> all the PEs.

Right.. But are you sure you want to policy-route traffic in your core
and thus don't want to label-switch IPv4 traffic?

Not sure what the requirements are, but some ideas come to mind:

1) advertise the (selected set of) destination prefixes within your IGP
instead of via IPv4BGP and don't advertise tags for them
2) Use MPLS TE and policy-route into the Tunnels on your PE's
3) Use L2TPv3 for RFC2547bis-VPNs and don't run MPLS at all (only
available in selected IOS/platforms, see
http://www.cisco.com/en/US/partner/products/sw/iosswrel/ps1829/products_
feature_guide09186a0080227c91.html)

Label-switching BGP prefixes within your core has the advantage of being
able to remove BGP from your core devices alltogether (obviously
requires that the core devices are pure LSR/P routers without any IPv4
customer or peering/upstream connection)..

	oli



More information about the cisco-nsp mailing list