The MPLS WG Archive

[Bora Akyol <akyol@pluris.com>]

Eric

I personally find the uniform model more useful than the pipe model. For
a transit backbone carrier, uniform model may make more sense than the
pipe model. The pipe model makes more sense VPNs, OXCs, etc.

Bora


Eric Gray wrote:

> Shahram,
>
>         Thanks for the clarification on use of DE/CLP.
>
>         Could you please move the paragraph that says
> "the Uniform Model allows LSPs to span Diff-Serv domain
> boundaries" (toward the bottom of page 12) to a spot
> closer to the point at which this model is defined (e.g.
> - 2.6.3).  This paragraph makes some things much clearer.
>
>         Let me see if I understand now.
>
> The reason why the penultimate hop needs to perform the
> EXP'->EXP conversion is because the next hop LSR cannot
> tell the difference between PHP'd packets and non-PHP'd
> packets therefore all packets have to arrive with correct
> EXP (or DSCP) value for the applicable LSP stacking level.
>
>         Is this correct?
>
>         I seem to recall that there is not a lot of current
> interest in LSPs that span service boundaries.  So is it
> fair to say that there may not be a need to support the
> Uniform model in the very near future?  Yet the Uniform
> model is specified as the 'base mandatory mode'.  The Pipe
> model is clearly simpler to support with or without PHP
> and the biggest operational difference is that it requires
> LSP tunnels to be bound by DS Domains - which is likely to
> be the dominant mode of operation in any case.  Can you
> provide additional clarification on this?
>
>         A big part of my concern is that any LSR is likely
> to need to support PHP. The Uniform model imposes additional
> configuration requirements on penultimate hops.  This is
> bound to increase the configuration required in the network
> as a whole.  While configuration is an easy out in specifying
> protocol, it's not usually a GOOD THING.  What's more, the
> specification should say what approach must be provided - so
> that, for example, the need to specify a standard MIB for
> configuring the required information is clear.
>
> --
> Eric Gray