The MPLS WG Archive

[David Charlap <David.Charlap@marconi.com>]

Markus Jork wrote:
> 
> It looks to me as if the L3PID signaling seemed like a good idea at
> the time but turned out to be not particularly useful. The problem
> is that it is often unknown what type of traffic is carried on an
> LSP.  For example an RSVP LSP can be used to carry IP traffic and
> at the same time also labeled traffic from a targeted LDP session
> that runs over the RSVP LSP.

The L3PID of the inner LSP is known, because it is signaled as IP.

The ingress router for the inner LSP also knows the L3PID of the outer
LSP, since it must participate in the signaling for the outer LSP.

I would argue that it should not shunt incompatible traffic into that
LSP.  Don't allow a non-IP LSP to tunnel through an IP-LSP.

If the inner LSP is meant to carry traffic of many different types, then
it should be signaled as such.  With an L3PID meaning "unknown".  Which
would mean, of course, that the egress of such an LSP would not be able
to handle a packet that only has one label - since it won't know the
actual L3PID after popping that last label.  But this would be OK, if
there's another label on the stack - either indicating an outer LSP to
use for forwarding, or indicating the L2/L3 PID.

Of course, the whole argument falls apart when you realize that the two
LSPs (inner and outer) don't both have to be RSVP.  I don't think L3PID
is signaled with LDP or CR-LDP.

-- David