The MPLS WG Archive

[Eric Osborne <eosborne@cisco.com>]

On Fri, Sep 29, 2000 at 11:30:25AM -0500, Chris Flores wrote:
> Eric - 
> 
> I agree. It is perfectly valid to tunnel packets through the core via MPLS,
> thereby avoiding a full IBGP mesh. In the steady state, we agree. I just
> wanted to point out that BGP is needed in the core in case MPLS fails for
> any reason. If the MPLS LSPs are not available, then the core will need BGP.


The statement "if MPLS LSPs are not available then forwarding will by
and large not work" is true.  But it's equally true to say ""if BGP
routes not available then forwarding will by and large not work".

This comes down to mainly an exposure issue.  Right now, people are
asking "what do I do if mpls fails and I don't have bgp"?  This is
generally because of either a lack of understanding *or* a desire for
belt-n-suspenders.  As deployments mature and operators get more
experience with them, there will be less need to back up MPLS with IP.

There's also the point (which you may get, but I didn't see you
acknowledge) that there are some MPLS services that are _not_ backed
up by BGP in the core.  If you want to run MPLS-TE, sure, it's nice to
be able to fall back on BGP in the core if things go wrong with
MPLS. Engineering around pathological conditions is up to the
operator, but not an entirely invalid idea.  But if you want to
provide MPLS-VPN services, or circuit transport, or other such fancy
stuff, then you will be unable to fall back to normal IP routing in
the core.



eric