The MPLS WG Archive

[Ben Mack-Crane <Ben.Mack-Crane@tellabs.com>]

Adrian,

In our work on MPLS recovery and protection switching we
recognized the following concern:

If the Notify is not constrained to retrace the path of the
LSP (upstream) and instead follows an IP layer forwarding path,
there is the possibility that it will run into the same fault
it is trying to report.  This could either cause it to be lost
entirely or delayed until IP routing converges (perhaps
much more delay than hop-by-hop forwarding in software).

What are your thoughts on this?

Regards,
Ben

AF@dataconnection.com wrote:
> 
> >>Note that whether intermediate nodes process the message is a
> >>distinct issue from whether the message traces the LSP
> >>hop-by-hop or goes along a completely different path.
> >
> >Care to elaborate on the importance of this issue?
> 
> Lou,
> I just want to separate the discussions.
> 
> I don't think it is a big deal if every LSR on the path of a Notify passes
> the message through software.  I just think it a shame not to offer the
> facility for a switch to fast-path the message (i.e. not remove it from
> forwarding hardware) if it is capable.  Thus I would be happy with Notify
> targeted at some remote node and router alert NOT set.  It is then up to the
> individual switch how optimally it processes the packets.
> 
> I believe it is important to allow a Notify to not retrace the path of the
> LSP.  This allows a Notify to be routed around a fialure (2nd order failure
> - so perhaps not too important).  It also allows a Notify to be sent to some
> other entirely distinct node (widening the purpose of Notify from simple
> protection switching).
> 
> Adrian