The MPLS WG Archive

[neil.2.harrison@bt.com]

Yakov,

> 
> Just to help more, here are few points from one of my 
> presentations on (G)MPLS:
> 
> (G)MPLS is not a Network Layer on its own, as it does not 
> have routing and addressing on its own - it uses IP 
> addressing + IP routing (with
> extensions)
NH=> You talk of addressing here Yakov and yes we can use IP (v4 or v6)
structured addresses all over the place....does not mean to say they
belong to an IP layer network however.  For example, I can use IP v4
structured addressing for an SDH VC4 access point and an SDH VC12 access
point (or the access points in any other layer network for that
matter)......but I for sure cannot connect these together as they belong
to different layer networks.  {A layer network is characterised by the
set of addressable access points which can be connected together.}

Addressing is bound to access points.  So even if folks use the same
addressing format they come from difference spaces.

Further, and as folks have discovered with GMPLS, or more accurately due
to the co-cs mode, the traffic data-plane network (co-cs) must be
disjoint from the control (or management) data-plane network (which in
all probability uses the cl-ps mode).  These 2 (at least) networks in
the same technology (say SDH) have quite different addressing
requirements.
> 
> (G)MPLS is not a Link Layer because MPLS works over various 
> Link Layer technologies (e.g., SONET, Ethernet, ATM, etc.)
NH=> (The control-plane of) GMPLS looks like it does because of the
co-cs mode.  The co-cs mode forces certain things on the network design:
-	there are no QoS classes in the co-cs mode
-	the control/management protocols must be run OOB wrt the traffic
in the co-cs mode....rather an attractive consequence for many reasons,
esp security.
-	the connectivity requirements of the co mode cannot be violated,
ie no mp2p allowed.....this also means the defects are different to
either the cl-ps or co-ps modes
-	the traffic data-plane must retain a fixed/known hierarchy in
the co-cs mode.  Now a word on MPLS is important here.  The 'shim' of
MPLS creates a digital wrapper behaviour in the data-plane of the co-ps
mode, ie there is no fixed/known hierarchy.  This is really neat in some
respects (esp OAM) but it does create some issues when we want to run an
MPLS network belonging to one operator over an MPLS network belonging to
another operator and retain full commercial/functional isolation.  This
is not a problem in GMPLS.....but its not because this was an a priori
design criteria of GMPLS, its a forced imposition of the co-cs mode.

So from a data-plane viewpoint GMPLS and MPLS are not at all alike.
> 
> (G)MPLS is not a Layer in the OSIRM (OSI Reference Model) 
> sense as it does not have a single format for transport of 
> the data from the layer above (e.g., (G)MPLS uses "shim" on 
> SONET, VCI/VPI on ATM, time slot on TDM, lambda on OXC, etc...).
NH=> The OSIRM is not a reflection of reality.  It has, however, led to
a great deal of confusion since folks think there is only ever 1
'network layer'.   There are lots of network layers (or more accurately
layer networks), and the only thing that roughly classifies them in any
sensible way is whether technology X belongs to the cl-ps mode, the
co-ps mode or the co-cs mode....and its is the differences in the modal
behaviours that are the important bits.  IMO functional convergence
within a mode makes a great deal of sense, but not across them (esp when
we need to take the commercial operating requirements of carriers into
account).

regards, Neil

_______________________________________________
mpls mailing list
mpls@lists.ietf.org
https://www1.ietf.org/mailman/listinfo/mpls