The MPLS WG Archive

[Giles Heron <giles@level3.net>]

Olle Pers wrote:
[snip]
> *  Your second point is key.  How can you know that the
> *  destination address in the packet is an IGP prefix?
> *  I suspect that many networks will want to run BGP
> *  edge-to-edge, but to have only IGP routes in the core.
> *
> The IGP knew the prefix, and B had sent out a label mapping
> for it. That was the scenario in David's question.
> So the router B knows how to forward unlabeled packets with this prefix. And
> it did send out a label mapping for exactly the same prefix, knowing that it
> didn't have any corresponding downstream label yet. Why? I don't want to
> think B did that just in order to drop these packets.

Yes, that is the situation in David's question.

However, as I see it the problem with allowing the router to attempt to
forward the packet unlabelled is that it will work in some situations
and not in others - so it is difficult to debug.  I have already seen
one situation where a router was forwarding traceroutes happily along a
"broken" LSP (and forwarding the packet as IP from the point where the
LSP broke onwards) but was unable to forward MPLS VPN traffic.  A
similar result would have occurred with Internet traffic being forwarded
over a BGP-less MPLS core.

> *  In addition to this you have the issue of MPLS VPN and Layer
> *  2 over MPLS traffic.  Either one of these will generate a
> *  label stack in which the innermost label is assigned edge-to-
> *  edge and is not visible to the intermediate devices.
> *
> These applications require that there exists an edge-to-edge LSP which can
> be used as a tunnel. We cannot assume them to work, before the setup of this
> LSP is completed. And with unordered control, the ingress will not know when
> this has happened.
> 
> But why use unordered control at all, if it leads to problems? What's the
> application?

My understanding of ordered control is that it leads to longer
convergence times than unordered/independent control?  If the
application is layer 2 circuit emulation one would hope to have
convergence times which are similar to those that can be achieved with
native layer 2 techniques (such as PNNI.)

Giles

-- 
===================================================
Giles Heron - IP Architect - Level 3 Communications
phone: +44 20 7864 0719     mobile: +44 7880 506185
===================================================