The MPLS WG Archive

[neil.2.harrison@bt.com]

Eric wrote:
	<snipped>
> Some  of the  feeling that  there are  layering violations  may stem  from
> a
> belief that MPLS  is a "layer 2" mechanism.  I've always  thought of MPLS
> as
> an extension  of the network layer.  Thinking  of it this way  makes it
> very
> natural to define it with  a combination of network-layer-specific rules
> and
> network-layer-independent rules.  If  the only set of
> network-layer-specific
> rules are those that apply to IP,  that simply reflects the fact that no
> one
> cares about the other network layers.
> 
	NH=> I don't agree with this.......and I really wish we could open
some minds here as to what is L1, L2, L3 or whatever....that OSI model has a
hell of a lot to answer for in terms of the blinkered thinking it has
induced in some, ie that there is only 1 L3 network and of course that is
IP.  There are lots of different layer networks (ie all operating at the
so-called L3).  MPLS itself creates new layer networks...this is a fact from
functional arch definitions.  Indeed, any layer network which can do
'find/route/connect' is L3.  With all due respect Eric, I think your
comments as regarding MPLS as an extension to IP is the root of many of the
problems people are expressing with MPLS....and I argue that if it were not
seen this way, then better progress would be made and more people would be
happy. 
	 
	Aside 1 - when layer networks are stacked (so lets say
OTN/SDH/ATM/IP for arguments sake, without being too specific about *which*
layer network within these technologies is considered, eg ATM has 2, VP and
VC layers) each has a full 7 layer OSI stack....its just that some of the
(above) layer (3) functions are null.  The only place that I have seen this
referenced is in the original book by Reid/Sexton (who were prime movers in
instigating functional arch models about 10-12 years ago) which dealt with
SDH/Sonet.

	Aside 2 - a similar case (ie opening minds) can be made for the much
hallowed 'end-end' principle.  End-end depends on one's perspective.  An
operator who sells SDH managed BW services will consider a VC4 (say)
'end-end'.  The only really meaningful definition I have for 'end-end' is
between the trail termination points of the layer network being considered
(see G.805, for example, for more info)....and of course, this also
correctly applies to IP wrt to the context it is applied, ie the points
where the IP OH is generated/terminated in the user-plane.

	neil