The MPLS WG Archive

[neil.2.harrison@bt.com]

Hi Maarten,

Good to see you looking at this stuff at last....it certainly needs some
functional arch rigour which you can bring.  Your points:
<snip>

> I get the impression that a LSP is being used for both the trail and a
> tandem connection (subnetwork connection).
	NH=> You need to consider what is happening in the user-plane and at
which level.

	For the user-plane at the packet level then a shim-header is used to
provide additional OH functionality.....so this is part of the trail term
stuff as you point out later.  And despite some reluctance to accept the
fact, this shim-header is creating new layer networks......some call it
'L2.5', but this is not accurate since all layer networks are L3 (you only
need addressing/routing to meet L3 requirements) and what is really being
talked about here is the difference in transfer mode, ie CNLS vs CO, of the
particular layer network technology concerned.

	However, once one considers what you and I would call transport
layer networks (eg SDH, PDH, OTN) the shim-header disappears as far as the
user-plane is concerned (why?.....because its no longer pkt-based), and all
that is used in the user-plane is whatever framing OH is defined for the
particular network layer technology considered.  So here, there is *no* new
trail term functionality.......the only issue being 'what control-plane
facets should be used?', and the aim here is commonality across all layer
networks (the so-called GMPLS)....which is OK *providing* the control-plane
facets are fit for purpose (and I am not sure some are, eg 32 bit addressing
in the OTN (irrespective of format, which is something else that need
studying) looks unacceptably too small to me).

	So, in the pkt-case we do have new layer networks and there is *no*
TC (subnetwork) entity....one can just create new nested layer trails to any
hirearchical depth one wants (this is really a digital wrapper if you think
about it)......whilst in the trad CO transport layer case there are *no* new
layer networks.  These are very distinct and different cases.  I tried to
point this fact out some time ago on the MPLS list....maybe you missed
it?......but I am still not sure everyone has got a grip on it yet.

	 But also as a tributary slot identifier. 
> 	Note - Tributary slot identifiers in other technologies are: 
> 	VPI, VCI, time slot, frequency slot (wavelength).
> 
> A tributary slot identifier is bound to a "link connection". 
> A signal passing through from an incoming link connection (via a "matrix
> connection" in a fabric) to an outgoing link connection may see a
> tributary slot identifier change. E.g. VPI_in=124 => VPI_out=765,
> timeslot_in=1, timeslot_out = 18, lambda_in = 15 => lambda_out = 63, and
> label_in=5354 => label_out=678521.
> 
> TTP and CTP are terms defined in association with the information
> modelling. 
> *	A TTP in the information model is equivalent with a Trail
> Termination
> function (_TT) and the common part of the Adaptation function (_A) in
> the functional model. 
> *	A CTP in the information model is equivalent with the Connection
> Point
> at the Connection function (_C) and the client specific part of the
> Adaptation function (_A) in the functional model.
> 
> A link connection (LC) starts at a CTP and ends at the next CTP:
> 	Link Connection: CTP - server layer trail - CTP
> 
> A matrix connection (MC) starts at a TTP/CTP and ends at the TTP/CTP at
> the other end of the fabric (connection function):
> 	Matrix Connection: TTP/CTP - TTP/CTP
> 
> A matrix connection is the smallest subnetwork connection (SNC).
> A subnetwork connection is in general defined as:
> 	SNC: SNC - LC - SNC, with the smallest SNC being a MC.
> 
	NH=> Agreed.  If I can paraphrase the key issue you are describing
above in func arch terms ........this is the differnce between 'relative
addressing' which is only interface unique and can change on a link basis,
as opposed to 'absolute addressing' which *must* be globally unique (at
least for the network access point population considered) and can only
change on a trail basis. [though there are subtle differences between the
'identifiers' used at a TTP and an access point which I don't want to get
into here]

> >  TTPs - are the IP interfaces/ports?
> 
> TTPs are the LSP termination points, where bits 20-31 of the MPLS label
> are added/removed and the information in the payload of this MPLS packet
> extracted.
	NH=> Only for the pkt layer technology cases is this true.  It does
not apply for trad transport networks, and the TTP here is detremined by the
framing OH used in whatever layer technology is considered.

> >  CTPs- are the labels?
> 
> CTPs are the boundaries of the link connection. The 20-bit label value
> is added/removed here. A link connection starts/ends at each LSR and
> LER.
> 
> Associated with a CTP may be a tandem connection TTP. Such tandem
> connection TTP adds another 32-bit MPLS label and associated MPLS OAM
> packet flow.
	NH=> This is where we disagree.....but I am open to arguments to
pursuade me otherwise.  I see each pkt-based LSP as a layer network in its
own right with *no* fixed hierachies.  CF ATM which has 2 clear hierachies,
ie VP and VC.  One cannot keep 'wrapping' labels in ATM (due to fixed VP/VC
structure) but one can keep wrapping labels in true MPLS.  These are quite
different cases and MPLS (user-plane at pkt-level) is quite unique in this
respect.  So I don't see there is a case for TC per LSP....one simply
creates a lower layer LSP (ie true server trail) over the subnetwork region
of the higher LSP one wants to look at.  I like this feature very much.

> >  LCs - are the connections/associations between 2 labels in two
> >  different LSRs?
> 
> A LC is the connection between the egress of one LSR [or LER] and the
> ingress of the next LSR [or LER].
	NH=> Agreed *but* only for a given LSP level where no LSP nesting is
considered, ie within a single layer network.

> >  SNCs- are the forwarding tables in the LSRs and LERs? (a connection
> >  between the TTP and CTP in the LER, or a connection between CTPs in
> >  the LSRs)?
> 
> The set of MCs (matrix connections) are the forwarding tables in the
> LSRs and LERs.
	NH=> Seems true.