Hey all - I've searched far and wide and can't find an answer to this ...

I have 300+ routers in the backbone area. I am wanting to move some off into areas of their own.

1 - Can I have a STANDARD area (0.0.0.1) with multiple connections back to the backbone?

2 - If the backbone ever breaks will it use that round about path if so?

3 - And does it really save me any CPU / convergence time / etc if all the routers in the STANDARD area still pull all the routes and send all their routes?

Thanks for your time!

PS - It seems like I had major issues with 6.45.6 and ospf peers... had to roll back to 6.44.6. The LSA table shows all routes but does not install them. Anyone else have those problems with a large ospf deployment and mixed versions?

I'm also curious about this.

We have (at least) two ABRs between backbone and subarea and it seems to work as expected.

You can definitely run an additional area and connect it at more than one point. This is why the backbone area exists and when there are redundant paths, they will be used.

Areas will definitely help with scaling OSPF when applied to areas of the network that aren't core transit. CPU should be a bit lower overall after area segmentation since you're not running the SPF calc on one large area and breaking it up into smaller flooding domains. Essentially each are runs SPF simultaneously and the routes get injected at the ABR / ASBR

However, at a certain size, it's easier to use iBGP along with OSPF and let OSPF map paths and BGP advertise subnets for hosts/customers.

Is stubbing and range/summary mandatory for take advantage of the benefits?

At what level (# of routers, # of routes) is an MPLS+iBGP solution a more logical to use than plain OSPF?

Is stubbing and range/summary mandatory for take advantage of the benefits?

It definitely helps. The topology information is what contributes to CPU overhead and thus convergence time. When you're summarizing, you're only sending routes and not the detailed area topology info.

At what level (# of routers, # of routes) is an MPLS+iBGP solution a more logical to use than plain OSPF?

There are two ways to look at this - one is to use an iBGP design from the beginning and be ready to scale immediately. The other is to migrate to iBGP once you've reached several thousand OSPF routes (where I typically see people start to have issues)


What kind of network are you running?

Is stubbing and range/summary mandatory for take advantage of the benefits?

At what level (# of routers, # of routes) is an MPLS+iBGP solution a more logical to use than plain OSPF?

MPLS LDP does not use BGP routes. So you still HAVE to use OSPF (unless you statically add labels I guess)

And LDP also does not work unless you have /32 routes. So summarizing can be used to suppress all the router-router links but you still want to distribute your loopbacks etc

We run a multi-area OSPF today and handling approx. 400 routes in total. The network is physically very "ring-like" and that unfortunally has a negative impact on convergance time.

We used to do area range/summary before, but that introduced very odd behaviour as going the wrong or sub-optimal paths sometimes. And sometimes it didn't work at all. Linknets in the backbone was always visible everywhere for some reason. We also use MPLS so stubbing and summarization has to be done carefully.

We recently looked at using OSPF+MPLS+iBGP setup, and use it on some places already. But it also introduces complexity.

@millenium7: Yes, we are aware of that.

How do you keep the customer routes out of OSPF and only announce them via iBGP? If I uncheck redistribute-connected I no longer get the loopbacks and link paths. Not sure how to use OSPF and iBGP and not have every route in every router still.

You need to use route filtering
I've said it multiple times, OSPF kinda sucks with its design limitations at scale in a WISP but it's what we have so we have to deal with it

Note also that additional areas can all be fringe networks but using the exact same area
I.e. there's absolutely no problem with every single router outside of 'backbone' all being area1 even if they are directly next to each other but with no link to each other. All routers that aren't connected all communicate through the backbone anyway, so it's not a design issue to simply re-use area1 or W/E number you want

If you need LDP you still have to distribute /32 routes, but you can suppress everything else with a default route

Yes you're going to need to split your network in multiple areas, a couple of months ago I was experimenting the same issue, all of my network (like 350 routers) were configured in a single backbone area, and the amount of LSAs was massive, that implied when a single path goes down, that update was going to be updated to the entire topology around the country, what we do is to isolate sections in a single area (default type worked for us); and we needed to replan our IP addressing because of each ABR was also summarizing LSAs as example, all of the WANs inside the area was in the x.x.x.x/24 subnet so the backbone area and other areas only learn a single LSA for your entire subnet making everything faster and reliable.