[c-nsp] OSPF design

[Brett Frankenberger]

On Fri, Oct 22, 2010 at 07:24:51PM +0700, Rin wrote:
> Hi group, 
> 
> We need to design a MPLS network that has around 100 nodes (7600) divided
> into Core, Aggregation & Access layer. OSPF and MPLS is deployed up to
> access layer. According to Cisco, an OSPF area should have no more than 50
> nodes in order to minimize the database. With that concept, we should design
> our network into different areas, say Core & Aggregation in Area 0, Access
> nodes in area 1. However, I reckon separating the network into different
> OSPF areas without summarization at ABR cannot minimize OSPF database, all
> routers still receive routes advertises by other routers. If we do
> summarization at ABR, MPLS cannot work since this is a continuous MPLS
> domain. 

You can't summarize the loopbacks[1], but you could summarize other
routes.  But, even without summaries, there are benefits to splitting
the areas.  It may not reduce the number of objects in the OSPF
database (in fact, in some cases, it will increase them, since stub
routes aren't carried as separate LSAs, but when they get summarized
into the backbone or into another area, they become individual Type 3
LSAs) ... but it reduces the complexity of the OSPF route calculation,
and it reduces the frequency with with the route calculation must run
(since it only has to run when there's a change in an area that the
router is in), and it creates opportunities for doing partial
reclaculation.

That said, I certainly agree with Oliver that 100 nodes is no big deal
with modern equipment.  (My largest area has 175 routers, most of which
have considerably slower CPUs and considerably less RAM than a 7600,
and it's full of microwave links, so there's a lot more topology
changes than you have in what I presume is a mostly fiber-based
network.  It works fine.)

> So my question is should we separate the network into different OSPF areas
> as Cisco recommendation or should we keep all routers in an OSPF area 0?
> Note that we intend to provide MPLS L2VPN, L3VPN on this network; we can
> only use OSPF, ISIS is not an option. 

One area.  (Probably even if you have customer routes in your IGP. 
Assuming you don't have stubby areas, if those routes are brought in as
externals, the Type 5's will be in every area regardless of how many
you have.)

     -- Brett

[1] For whoever asked, the reason for that is that MPLS VPN relies on
there being a /32 route to to the loopback of each PE so that there
will be a distinct LSP to each PE.