The MPLS WG Archive

[Curtis Villamizar <curtis@fictitious.org>]

In message <08d301c4395a$29b56490$67849ed9@Puppy>, "Adrian Farrel" writes:
> > > Perhaps someone could usefully define (precisely!) soft and hard pre-empt
> ion.
> > > 
> > > (Hint I thought I had done this, but perhaps a discussion on the list wou
> ld b
> > > e useful.)
> >
> > One of the neat things about this running code thing is that what ends
> > up really counting is who wrote some running code or worked to deploy
> > that code in a network particularly when the majority of vendors have
> > done the same thing and they all interoperate.  That's the case here.
> > 
> > I agree with you that we should clarify RFC3209 and preferably put the
> > clarification in a future iteration.  Perhaps it would be easier if
> > you suggested wording for the next iteration of RFC3209 and if the
> > authors agree to it, we won't need a separate document.  But if you
> > feel that's not the way to go or insufficient, go ahead and write an
> > informational draft with the clarification and a crystal clear
> > definition of hard preemption and soft preemption.
> 
> Clearly someone with running code needs to:
> - document what is meant by soft/hard pre-emption
> - document what the running code does.
> 
> Over to you, Curtis.
> 
> Adrian


Adrian,

As your survey showed, although there were differences in exactly what
each router did, these differences were of no consequence as long as
the ingress made the assumption that the path was no longer suitable
to forward over and rerouted the LSP to another path on any Path-Err.

Therefore an adequate definition of hard preemption is any preemption
in which the forwarding along the entire path is no longer known with
certainty to be intact.  By this definition, most or all known
deployed RFC3209 implementations do hard preemption.

Soft preemption is a condition where for a given LSP, forwarding is
known to be in place along the entire path for some finite and
non-zero grace period, however at one or more places resource limits
have been exceeded and that LSP has been identified for elimination.

Now all we need is a mechanism which meets the above criteria for
"soft preemption".  I think Mathew has such a mechanism already
defined.

Curtis