The MPLS WG Archive

[neil.2.harrison@bt.com]

> Curtis,
> 
> > 
> > LSP Ping does not require an RSVP-TE reverse path.  The ICMP echo
> > reply is delivered to the IP address of the sender.  Its and IP
> > packet.  It doesn't need an LSP but if one is there it can use it.
> > 
> > Curtis
> > 
> 
> I think you need to read the LSP-Ping draft one more time. I 
> didn't say that
> The ICMP echo reply needs RSVP-TE reverse path, rather the 
> LSP-ping response requires
> RSVP-TE reverse path. The reason is that it needs a disjoint 
> path from the normal
> IP hop-by-hop path, in order to determine that the ICMP-echo 
> response path wasn't at fault.
> 
> -Shahram
NH=> Yes Shahram, but even this can't be assured can it, ie that IP and MPLS
layer return paths are disjoint?  If they are not, what do we learn?  That
either the go or return paths (or both) are dead (I am ignoring other defect
mechanisms that any pinging scheme would fail to recognise and act on
anyway).  And in the meantime whilst we scratch our heads (and probably
before the pinging even began), alarms could be ringing in client networks
in other domains/countries.  Bottom-line in all this is:  if fault detection
is going to be simple, unambiguous and ensure correct actions are taken
exactly where needed it has to be unidirectional in nature.  IMO I find it
really hard to see how anyone could seriously argue against the obviousness
of this statement.