The MPLS WG Archive

[Kireeti Kompella <kireeti@juniper.net>]

LEI YAO wrote:

> Also, the label is locally significant. There is no way for the upstream
> LSR to know and suggest label value for the downstream LSR.

Actually, there is.  Locally significant means that both ends of a link
know what's going on, but others LSRs in the path don't.  Suggested labels
work fine.

Note that the above is true for point-to-point links only.  On multipoint
links, suggesting labels doesn't work as well, as one could have label
conflicts.  Fortunately, non-packets links are (to date) point-to-point.

Yangguang wrote:

> Here is the difference between circuit switches and data LSRs. An circuit switch
> builds up label association with its neighbor at the system initiation time. The
> port association table is the label association table (not exactly). It's known
> before connection time. Please refer to ietf-xu-mpls-ipo-gmpls-arch-00.txt for
> details.

I don't get it.  Do you mean using LMP or similar mechanism for
establishing the mapping?

If a "data LSR" tells its neighbor to use label 272, the neighbor
knows exactly what is meant.  If a SONET LSR tells its neighbor, use
label foo, where foo is an encoding of the time slot to use, the
neighbor knows exactly which time slot to use.  However, if a PXC tells
its neighbor to use label bar, the mapping between "bar" and port number
has to first be established.  The distinction is not data and non-data,
but defined semantics for labels vs. dynamically established semantics.

Kireeti.