The MPLS WG Archive

[Robert Raszuk <raszuk@cisco.com>]

> And this same nightmare doesn't exist if you provision hundreds or
> thousands of LSPs through that exact same network?

When you provision let's say TE tunnels (in not all but in the most cases) you
are not building them from your customer interface to customer interface on the
other end of your network. What you are building  is a set of tunnels to
interconnect your own boxes with (either edge-egde or middle-edge or any other
possible combination) but always within you domain. That itself reduces the
number of LSPs versus VCs very dramatically. In MPLS-VPN by using for example
prefix based label stacking you don't need to provision any LSP/VC to create
very large VPNs across your core.

Of course one may argue that instead of building L3 networks with IGPs just
flat or hierarchical PNNI should be used in a closed provider network to achive
the same. Unfortunately by looking at the current industry trends, dominance of
IP and proven scalbility of BGP & IGPs in the Internet I don't think that this
option is practical.

R.

> David Charlap wrote:
> 
> Robert Raszuk wrote:
> >>
> >> All of these things are things that exist and can be supported using
> >> ATM (for example).  Yet people you're talking to want them using
> >> MPLS.
> >
> > That is true that all of them can be provisioned with ATM today. But
> > the problem is that using L2 (or any kind of tunneling) for TE, for
> > VPN or even for Circuit Transport scales very poorly. Once you
> > provisioned hundred or maybe thousand of VC your operations people are
> > starting to suffer from lack of control and maintenance nightmare.
> 
> And this same nightmare doesn't exist if you provision hundreds or
> thousands of LSPs through that exact same network?
> 
> -- David