The MPLS WG Archive

[neil.2.harrison@bt.com]

Hi Yakov,
> being practical kind of guy, I have no interest in 
> theoretical arguments.
NH=> Ditto....but we sometimes need them to make a more general point
> 
> > If you also take your analogy further to
> > GMPLS, then perhaps we should observe that SDH/Sonet never 
> really existed
> > until they were blessed with the IP-based control-elements?  
> 
> SDH/SONET certainly existed before GMPLS. But that has nothing to do
> with the *fact* that GMPLS is a control plane that uses IP 
> addressing/routing/signaling (and *not* PNNI).
NH=> Agreed....but the point is their user-planes do have the required
fault-handling behaviour that can exist irrespective of the
health/behaviour/existence of the (GMPLS) control-plane...I know because
we/others put them in there.  This is actually quite a good principle in CO
networks...and it has a corollary in that if the control-plane dies or
misbehaves, it should not affect the user-plane (which we should know is
'good').
> 
> > MPLS does create layer networks in the user-plane, GMPLS 
> does not....these
> > existed before.  
> 
> The discussion on whether MPLS does, or does not create layer 
> networks,
> seems to be of little/no practical significance.
NH=> If you agree we have a new forwarding paradigm with MPLS.  If you agree
we have new forwading plane functionality (like the label....not a globally
unique entity like the IP address) then it follows that these 'components'
can be subject to failure mechanisms that did not exist before with IP only.
It therefore follows we have to detect/handle these.  Since the LSP can
exist outside of any specific routing or signalling protocol then we have to
ensure control-plane decoupling.  We should *never* have any specific client
or server coupling for add/change/remove of layer networks (same actually
goes for the control-lane protocols).  Ergo MPLS needs its own user-plane
fault-handling mechanisms for quite practical reasons.
regards Neil