The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2004-Apr> msg00013



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

MPLS-TE questions

  • From: David Charlap <David.Charlap@marconi.com>
  • Date: Fri, 02 Apr 2004 10:05:15 -0500
  • Organization: Marconi, Vienna VA
  • User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.6) Gecko/20040113

Alexander Attsik wrote:
> I need to understand some principles of MPLS-TE, could you help me please.
> 
> In MPLS-TE, RSVP-TE (or CR-LDP) sets up explicit routes, previously computed
> by means of Constraint-Based Routing. In light of this I have following questions:
> 
> 1. Who sets the above mentioned constraints and initiates the procedure of
> Constrained Shortest Path calculation?

This is beyond the scope of the signaling protocol (RSVP-TE or CD-LDP). 
  These parameters will be given to the signaling protocol on the 
ingress node by some external source.

This source may by a path-computation engine on the ingress node, manual 
operator configuration, and off-line management tool, or anything else. 
The signaling protocols don't know and don't care where this request 
comes from.

> 2. And quite similar question:
> 
> Imagine such situation:
> 
>  A----(B---- . . . ----C)----D
> 
> B and C are edge routers of MPLS-TE domain
> A and D belong to non-MPLS  areas.
> 
> A wants to communicate with D, is it possible for A to ask B for some QoS
> level and thus initiate
> TE-LSP setup from B to C? How can A do this (I suppose, by means of RSVP
> (rfc2205)).

RFC 2205 is designed for application-to-application reservation of 
resources.

MPLS-TE is really meant for router-to-router allocation of resources. 
It would, of course, be possible for applications to initiate LSP setup, 
but so far, I have not heard of anybody planning to release some kind of 
"mpls to the desktop" system.  (This alsoc changes the definition of 
your problem, since in a to-the-desktop scenario, A becomes the edge 
node for the LSP.

Typically, an application will just send unlabeled packets and be 
unaware of any MPLS.  When those packets hit the first LER (B in your 
case), that router will classify the packets and forward them into the 
approrpriate LSP.

Although it is theoretically possible for B to auto-create the LSP in 
response to data (in much the same way that ATM LANE can auto-create VCs 
in response to data), I haven't heard of implementations actually doing 
this.  It is more likely that the network operator will have established 
this LSP in advance.  If the LSP is not there, router B can either 
forward the traffic hop-by-hop (assuming that the rest of the network 
has appropriate routing tables in place) or drop it (if the network 
can't forward the unlabeled traffic.)

-- David