[c-nsp] MPLS - MP-BPG with multiple OSPF areas

Livio Zanol Puppim livio.zanol.puppim at gmail.com
Mon Oct 24 20:37:39 EDT 2011


Thank you very much for your answer.

Searching the Internet about this topic, I've found a draft that seems to be
a near future solution for some major MPLS networks, called "seamless MPLS".
As written in the laymann draft (
http://tools.ietf.org/id/draft-ietf-mpls-seamless-mpls-00.txt), the
implementation uses 2 levels of IS-IS (or 2 OSPF areas).

Also, pseudowires and LDP downstream on demand are used. I know it's only a
draft, but I find it usefull.

Anyway, thank you for your advices.

2011/10/20 Mark Tinka <mtinka at globaltransit.net>

> On Thursday, October 20, 2011 06:39:38 AM Livio Zanol Puppim
> wrote:
>
> > About MPLS features in multiple areas, aren't there
> > solutions that can be implemented to overcome the
> > limitations like LDP prefix leaking or some solutions
> > with mGRE?
>
> The IGP already knows about the individual infrastructure
> addresses between areas, which is what helps LDP to work.
>
> But MPLS-TE is something else. Paths have to be signaled,
> and by default, ingress routers signaling paths tend to
> prefer that the tail-end of the LSP be in the same area or
> level. If that isn't the case, Inter-Area TE is required.
>
> > What issue do you see keeping a lot of routers in a
> > single area (let's say 400) versus routers in different
> > areas?
>
> In OSPFv2, Areas help the routing protocol to scale,
> primarily due to the close ties between Type 1 and Type 2
> LSA's. This isn't an issue in OSPFv3 or IS-IS, as Topology
> and Reachability information is separated.
>
> At the risk of starting a routing protocol war, IS-IS has
> been known to scale very well in single level environments
> (see Vijay Gill's OSPF-to-IS-IS migration at ATDN).
>
> We have two networks, one where we run IS-IS in one level
> (L2) and one where we use multiple levels. It scales very
> well in both, and both are fairly large.
>
> There are rumours that OSPF can scale to some 10,000 routes
> on today's kit, but I'm not sure as I don't run it.
>
> > Are there any recommendations about that or about the
> > maximum number of routers in a single area, or maybe
> > something about benefits/problems in both designs?
>
> The problem with a single area/level is that one router
> sneezing a million miles away is heard by all other routers
> at the topology level. For such a problem, OSPFv3 or IS-IS
> will react better.
>
> Areas/levels help to scale the network, but introduce issues
> like MPLS-TE. In our IPTv network, where we use NG-MVPN for
> Multicast (p2mp RSVP-TE), we use a single L2 IS-IS domain,
> just to avoid issues or configuration complexities with
> signaling p2mp paths across different parts of the country.
> The risk is increased noise, but the kit can take it.
>
> Mark.
>



-- 
[]'s

Lívio Zanol Puppim


More information about the cisco-nsp mailing list