The MPLS WG Archive

[Curtis Villamizar <curtis@avici.com>]

In message <00f801bf9843$888c2560$425cc697@tempest>, "Krishna Bala" writes:
> Jagan,
> Just to beat a dead horse .. the difference between the open model
> and the overlay is just that the overlay model typically suggests that
> there would be N^2 connectivity (i.e., every router, for instance, is
> statically connected to every other). In any case, the open model allows
> for fully dynamic operation. The overlay model suggests static operation of
> the optical layer.
> 
> As I see it there are only two architectures that are worth considering:
> 1. Peer
> 2. Open
> 
> I agree with you that there is no need to add the "overlay" model to
> this list. The open model covers the static and the dynamic cases.
> 
> Krishna


In terms of visibility into the optical layer and control over path
selection, the proposals on the talbe can be characterized as follows:

  1.  Static Overlay Model - paths endpoints are specified through a
      network management system though the paths may be laid out
      statically (by the NMS system) or dynamically (by the network
      elements).  This is similar to ATM PVCs and SPVCs.

      [Note: networks built using the overlay model need not provide a
      full mesh contrary to misinformation provided on this mailing
      list.  Today ATM PVC based IP backbones are often not built as
      full mesh networks to reduce the nuber of routing adjacencies.]

  2.  Signaled Overlay Model (aka "Open" model) - paths endpoints are
      specified through signaling via a UNI.  Paths must be laid out
      dynamically since they are specified by signaling.  This is
      similar to ATM SVCs.

  3.  Peer model - The network elements are provided enough
      information such that they MAY attempt to set up path in
      addition to specifying their endpoints.  Paths may be laid out
      by the optical elements if the edge LSR provides only a static
      hop.  Paths may be laid out by the LSR.  Paths may also be laid
      out statically by having the NMS system configure explicit paths
      on the LSR and have those paths signaled.

It was my understanding that the MPLS WG traffic engineering work was
motivated largely by major Internet service provider's experience with
ATM and was making an effort to avoid repeating the mistakes of the
ATMF and therfore (among other differences between ATM and MPLS) is
interested in the peer model.  

Note - If the OIF wants to ignore the experience that the Internet
community has had with ATM and wants to ignore the feedback through
the IETF of people involved in building those networks and and build
IP routers with ATM interfaces that were constrained to make use of
the interfaces that the ATMF chose to provide (in some cases omiting
capabilities over the loud objections within IETF WGs such as IP over
ATM) then that is a matter for OIF.

If we are putting together a framework, then in addition to the
"static overlay model" and "peer model", a distinction should be made
and a separate "signaled overlay model" can be provided.  In the
interest of completelness the framework can point out that the OIF is
working on a UNI which is based on the signaled overlay model.

I prefer the term "signaled overlay model" over "open model" because
the former is descriptive and that latter seems motivated by the
desire to use the marketing buzzword "open" rather than provide a
descriptive term.

Some people had envisioned going directly from static overlay model to
peer model or going directly to the peer model with the understanding
that the LSR can still be configured with explicit paths from an NMS
as long as quantifying and signaling impairments remains slightly
beyond the state of the art and determining the impact of optical
impairments remains a black art.

Curtis