The MPLS WG Archive

[Curtis Villamizar <curtis@workhorse.fictitious.org>]

In message <3A5B8845.56E6E7D4@nortelnetworks.com>, "Cheng-Yin Lee" writes:
> Curtis,
> 
> Do you agree that :
> 1) the scaling issues in IGPs, are a result of fully-meshing routers, not a
> result of using independent control planes

Yes, but independent control planes was the reason dense meshing in
the IGP was needed.  The IGP had to run over the VCs.

You can have a full mesh of MPLS LSPs and a sparse mesh of IGP
adjacencies.  The same is not true for ATM.

> 2)  service providers who have requirements to fully-mesh routers  (for simpl
> e
> edge to edge traffic measurement, bandwidth management, control, as described
>  by
> Vijay)  have to setup fully-meshed circuits,  VCs or LSPs. The needs of  serv
> ice
> providers to fully-mesh routers do not arise because of ATM (or MPLS).

You don't seem to understand that with MPLS the mesh of MPLS LSPs is
decoupled from the IGP adjacencies.

With MPLS, the IGP does not run over the MPLS LSPs.  The IGP runs over
the physical links.  The IGP and TE mesh are decoupled.  You can have
BOTH a full mesh of LSPs AND a scalable IGP.

With ATM, the IGP runs over each VC.  The IGP and TE mesh are coupled.
You can have a full mesh of VCs OR a scalable IGP, but NOT BOTH.

> thanks,
> cyl

I'm repeating myself above in hopes that you'll finally see the point
that we've been trying to make.

Curtis

> Vijay Gill wrote:
> 
> > This, along with some hierarchy, allows for precise traffic measurement
> > between router-router flows and at the city-pair level, and results in
> > fairly tractable flow data that can be used to predict traffic levels,
> > allowing for proper sizing of sPVC contracts, etc.  These contracts are
> > then used by the ATM switch software to do constrained based routing of
> > sPVCs, allowing the network operators to do a good job of traffic
> > engineering and circuit layout/capacity planning.
> >
> > This might explain why people have gone with the extra expense (both in
> > capex and SGnA) and complexity to implement a fully meshed ATM solution.
> >
> > /vijay