RE: decisions on L2 solutions documents

On Tue, 2003-06-03 at 12:44, [email protected] wrote:
> Giles,
[snip]
> > Meanwhile, back in the real world, PWs (in the MPLS context) do the
> > following:
> > 
> > 1) add a label to the stack to keep state out of the core 
> > (this may also
> > create a p2p LSP on top of an mp2p LSP, depending on whether you are
> > using LDP or RSVP-TE for your tunnel LSPs).
> NH=> This is an LSP issue and belongs to the proper MPLS layer
> network....nothing to do with PWs.

of course there is no such thing as an MPLS layer network.  In an
architecturally pure parallel universe perhaps we'd have one ;-)  In any
case please let's not have that debate again...

but sure, label stacking is part of the MPLS architecture.   In this
case the PW label is signalled by the PWE3 control component.

> > 2) create a bidirectional p2p construct out of two unidirectional p2p
> > LSPs.
> NH=> This is an LSP issue and belongs to the proper MPLS layer
> network....nothing to do with PWs.

last time I checked MPLS LSPs were unidirectional, and PWs required
bidirectional connectivity.

> > 3) adapt the layer 2 protocol being carried to MPLS.
> NH=> Now you have it....PW are about adaptation....with that I agree.
> 
> > This involves
> > three things:
> > i.   stripping off the layer 2 header (aka "compression") in the cases
> > such as FR where the header may be different at each end.
> NH=> Yep, bit of BW squeezing

BW squeezing is a side-effect of stripping the L2 header, not the
primary reason it was done AFAIK.

> > ii.  sequence numbering, if required, for the case where the core
> > network may reorder packets.
> NH=> Yep, there to fix lower layer MPLS problems.

not really.  It's just that some client protocols may be less tolerant
of re-ordering than others.  So some SPs may want to have the option of
building a network which does not guarantee strict ordering of packets
in all circumstances, and then have the adaptation function provide this
for any traffic which needs it.

> > iii. length checking, for the case where the core network's 
> > minimum PDU
> > is larger than that of the PW (this is required because MPLS has no
> > length field).
> NH=>Agreed, adaptation function should pad to minimum.

at least we're agreed on something ;-)

Giles

-- 
=================================================================
Giles Heron    Principal Network Architect    PacketExchange Ltd.
ph: +44 7880 506185              "if you build it they will yawn"
=================================================================