A simple OSPF question related to NBMA and split Horizon

[aravindhs]

Hi !

Assume there are three routers - A, B and C.

A is the Hub. B and C are the spokes.

Each of ther routers A, B and C have a LAN interface in fa0/0 and a WAN interface s0.

On router A, there are 2 FR PVCs on S0. A to B and A to C.

There is no PVC between B and C.

The OSPF network type on S0 is point-multipoint.

The s0 interfaces on A, B and C are in the same subnet (the typical config for a multipt i/face)

Now, OSPF has 2 neighbors on s0, ie B and C through which it learns the LAN subnets of B and C sites.

My question :

==============

We know if this physical setup was for EIGRP, B would not be able to reach C because of split-horizon (same interface).

I am assuming that OSPF does not have any split horizon since it is a link-state routing protocol and hence B's LAN subnets learnt via the pvc to B would be flooded out in the form of LSAs to C and vice-versa.

I have not been able to try this out yet..

Expert advises please .. ?

Thanks in advance !

Arav

[ruwhite]

This depends on how you have the interface configured on the hub router (and the other routers). If you have it configured as: "ip ospf network point-to-multipoint," then the remotes will be able to reach each other. OSPF will actually inject host routes for each remote site's next hops, so the other remote sites will have a good next hop through the righ PVC to be able to reach those other remote sites.

In other words, A will inject a host route to C's s0 interface address, and advertise this route to B, so that B will mark it's next hops for networks behind C as being reachable through A, making the connectivity work. If you're using a broadcast type, or non-broadcast type, then you'd have to configured these /32's to make intersite connectivity work manually.

BTW, EIGRP turns off split horizon on point-to-multipoint links automatically, so site-to-site connectivity works there, as well.

:-)

Russ.W

[aravindhs]

Thanks Russ.

What you said about the host routes (/32s) is spot on!

My concern actually was whether the remote sites B's Lan subnets that the router A (the hub) learns through s0 would be sent out to C through the same interface s0.

I checked it out in the lab (with all possible debugging turned on )and found that there is no such thing as 'split-horizon' with Link State prototols (ospf in this instance)..

So, apart from the /32 routes for each spoke, the LAN subnets were also flooded out through Type1 LSAs (all in the same area).

Also, thank you Russ, for reminding about the EIGRP tip (auto split-horizon turn-off on a mpoint link.. )

Would be nice if anyone else can share any particular interesting experience related to this ..

Cheers

Arav