The MPLS WG Archive

[Curtis Villamizar <curtis@workhorse.fictitious.org>]

In message <3.0.5.32.20010406162005.0081bd20@10.1.1.249>, Mark Duffy writes:
> Hi Bill,
> 
> Your comment raises another question which I have been wondering about.
> RFC 3036 sect. 2.1 gives very specific rules for what *packets may be sent*
> on a particular LSP.  But it doesn't say anything about what higher-level
> *LSPs may be stacked* on the LSP. 
> 
> Presumably one stacks higher level LSPs whose next hop matches a FEC
> element associated with the outer LSP.  If one applied the rules of section
> 2.1 to the stacking, it would (for example) require that an outer LSP with
> a Host Address FEC be chosen in preference to any LSP with an Address
> Prefix FEC.


For most usage of MPLS, the notion of FEC is a broken one.  In
RSVP/TE, there is no FEC signalled which is a good thing.  For a
hierarchical tunnel, you can carry any traffic that needs to pass that
way.  A specific case would be a BGP VPN tunnel directly inside a TE
tunnel (which elsewhere could go directly inside another TE tunnel).

If an outer tunnel will only accept inner tunnels with a specific
L3PID, then we set that L3PID on the outer tunnel.  If it is allowed
to carry any L3PID, then the MPLS L3PID is used.  If the MPLS L3PID is
used, this also indicates that there must be another label underneath
since there is no way to determine how to interpret the payload.

Curtis