Multiple ABRs connected to Area0 and Area1 concern

[Brett Martin]

Hello,

I am interfacing 2 ABR's to a microwave management network configured as Area1 and all of the nodes in Area1 are configured as stubby. The fact that there are multiple ABR's, is there any concern with routers in Area 0 seeing routes to other Area0 networks via the Area 1 network? Basically I am concerned that Area 0 traffic may traverse the  Area 1 stub network (area0-area0). My understanding to this point is that a stubby area (area1) will not advertise any other areas so does this mean that Area 1 will only see default routes to Area 0 but Area 0 will not see Area1 as an option to route Area0 to Area0 traffic?

Thanks,

Brett

[Rolf Fischer]

Brett,

probably you already know that OSPF uses distance-vector mechanisms between areas, so there have to be some special loop-prevention mechanisms as well, called "inter-area loop prevention".

The most important rule to answer your question is that an ABR does not accept summary LSAs from non-backbone areas (others than area 0). So your ABR-2 will ignore the summary LSAs (prefixes originated in area 0, translated by ABR-1) received from ABR-1 through area 1. Result: No re-translation back into area 0. (This is, by the way, one reason why all areas have to be connected to area 0).

There is a great document with many details written by Petr Lapukhov: Inter-Area Loop Prevention in OSPF

Feel free to ask for further information.

Hope that helps

Rolf

P.S.: I left aside the special area types like stub etc.; they just filter certain LSA types for efficiency but not for loop prevention.

[Brett Martin]

Thank Rolf for that great explanation. To further understand this, I have another question I was hoping you could answer.

When Area1 hears LSA's from the 2 ABR's, because Area1 is "stubby", will it only advertise networks that are part of its 'stubby" area back to the ABR's? I assume so. If that is the case, would these 2 ABR's not receive any LSA's that have originated from them toward the "stubby" network as the stubby edge routers would filter them out?

Secondly, Area1 has approximately 15 router nodes. The ABR routers connect somewhere in the center of this star network and are separated by only one Area1 node. How does the stubby network resolve the path back to Area 0 with both ABR's advertising networks and default routes. I would assume based on the advertisements/LSA's from the ABR to the stubby network it must contain a metric that would be used by the stubby network devices. Is there a way to create a "preferred" ingress point to Area0 from the stubby network perspective?

Thanks in advance,

Brett

[Rolf Fischer]

Brett, you're welcome!

I'm not sure if your area 1 is a Stub Area (area stub) or a Totally Stub(by) Area (area stub no-summary on the ABRs).

What the ABRs in those area types do, is filter out certain information comming from the backbone and replace them by a default-route. This is done to save resources (CPU, memory, network traffic), but the price for that savings is loss of detail-information. Depending on the number of exit-points and the topology, this could result in suboptimal routing for some of the prefixes. Choosing the adequate area type is always a design question too.

Inside an area (no matter what type of area), all routers synchronize their link-state databases (LSDB) and once this is done, they all have the same topology information to calculate the shortest paths to the destinations, which then are installed into the routing table.

This is also true for ABRs, but they have a LSDB for every area they have an active OSPF interface in.

ABRs now translate the potential routes learned from OSPF (form the RIB, not from the LSDB!) from the one area into summary LSAs for the other area, the resulting routes are called "inter-area" routes ("O IA" in "show ip route"). Again, this new information is synchronized within the whole area.

In a stub area, the ABRs filter out external OSPF information (Type-5 LSAs) and replace them by a Summary LSA for a default-route, pointing to the injecting ABR. Again, the routes inside the stub area learn this default-route after the LSDBs are synchronized.

A totally stub area goes one step further and let the ABRs filter the summary LSAs (other area's "OSPF routing-table" information) too. So the only thing that an internal router in a totally stub area knows about the outside OSPF world are default routes injected by the ABRs. All the OSPF network prefixes from inside the (totally) stub area are learned nevertheless; the default route is needed for the outside only.

Normally an internal router then will choose the ABR with the lowest path cost (this is something you can control by configuration) as the exit point, you can check this with the "show ip ospf border-router" command.

There are some exceptions and special rules but I don't want to overcomplicate things - this already has been a lot of information, I guess ;-)

Hope that helps

Rolf