[c-nsp] MPLS network on 3750 switches - ISIS or OSPF which is scalable?

[Oliver Boehmer (oboehmer)]

Vikas Sharma <mailto:vikassharmas at gmail.com> wrote on Monday, October
15, 2007 10:59 AM:

> Hi,
> 
> I have approx. fifty 3750 switches and I have to implement MPLS
> network on that.

You are talking 3750ME, right? "regular" 3750 don't support MPLS.

> I am planning for OSPF in a single area as there
> will be only loopback IPs and connected routes in global IP routing
> table. But I am not sure abt he LSA flooding as my network is a full
> mesh. Though I can use database-filter command but to configure this
> command on every router is cumbersome. 2nd though is to implement
> ISIS with L2 level across the network. I want to understand which is
> more scalable with the kind of 3750 switches, ISIS with level 2 or
> OSPF with area zero?       

are you really sure you want to do a 50-node full-mesh network? Are you
talking about 49 p2p circuits per node? How do you plan to build them?
As far as I know, the 3750ME supports MPLS (i.e. label
imposition/disposition) on the two ES ports only. You want to use
subinterfaces here? How do you want to scale this in case you run out of
ports/capacity? Can't you use some hierarchy?

I don't have personal experience with a 50-node full-mesh Link state
network, but I feel you need to restrict the flooding using ISIS
mesh-groups or OSPF database filter. Both are cumbersome as you need to
find a compromise between flooding surpression and robustness. Both ISIS
and OSPF should scale in a similar fashion, so don't think it makes a
difference which one you'll choose.

I would, however, re-evaluate the flat full-mesh design and think about
some hierarchy.

	oli