The MPLS WG Archive

["Adrian Farrel" <afarrel@movaz.com>]

Curtis,

> The current semantics of path-err or path-tear is to indicate an error
> in setup or to tear down.  Any intermediate node that didn' understand
> this would release resources and cause traffic to black hole, exactly
> what the soft preempt is trying to avoid.

Interesting, but I don't believe this is the semantics of a PathErr. In fact,
quite to the contrary. This is why RFC3473 introduces the Path_state_removed
flag.

To quote from RFC3473 section 4.4
   The PathErr message as defined in [RFC2205] is sent hop-by-hop to the
   source of the associated Path message.  Intermediate nodes may
   inspect this message, but take no [sic] action upon it.

That is: the existing PathErr preemption scheme is *already* soft with two
exceptions...
- the preempting node currently releases the resources of the LSP
   (this is local behavior and can be modified by implementations)
- the ingress node currently sends PathTear
   (this is local behavior and can be modified by implementations)

We might want to add further local behavior to "allow" a transit node to inspect
and act on a preemption PathErr.

None of these changes speak to the need for a signaling change, but it might be
useful to indicate how the preempting node has behaved.

> The soft preempt is an advisory from a midpoint to the ingress that
> while an LSP is up, its resource reservation cannot be maintained and
> it must be voluntarily rerouted or face tear down.  A bit in the
> normal resv is a better way to express this than one of the error or
> explicit tear down mechanisms.

I disagree as above because PathErr is not an explicit tear down mechanism.

> Also, any midpoint that recieves this and then has to do a preempt can
> prefer to soft preempt an already soft preempted LSP over one that has
> not yet been preempted.  This would be quite common during the
> reaction to a link failure, if the rerouting of preferred LSPs caused
> less preferred LSPs to be preempted.

I believe this is facilitated equally by a PathErr.

Thanks,
Adrian