Split Horizon Issue

[h.bradley]

I have 55 routers connected into an ATM/radio network. I am running EIGRP. This network basically acts a bridge. All Router WAN ports are on the same subnet. There are 6 main locations on this network with up to a dozen of my routers hanging off them. These main locations are meshed to all the other locations so that the routers are free to talk to any of the other locations. The big problem is that because of the amount of eigrp information flying around, in conjunction with ATM PVC traffic, the most optimum path to other locations keeps changing. This results in timeouts for user apps, etc. I've adjusted the eigrp hello interval and the hold timers, but this hasn't helped matters any. I could get the carrier to remove the meshing, but split horizon would prevent the sites from being able to talk to one another (Each of the 55 sites would come back to our main WAN router at the head office). Removing the meshing and static routing everything would work, but would be an administrative nightmare. All documentation that I have read says that disabling split horizon is not advisable.

Any ideas?

Regards,

Hugh

[ahbanks]

Some clarification, please..

Are these PVCs or SVCs?

How many EIGRP AS do you have?

How reliable are your links, are they all over radio, or just some?

Do your 6 main locations advertise summary addreses, or are all routes sent to all locations?

Offhand, it sounds like a link flapping. BGP can handle this more elegantly, with its hold-down facility, but that may not be an option in your topology.

[smif101]

Can you provide a logical and physical diagram of your network including IP addressing and ATM PVC's. From what it sounds like you though you should build a heiarcheal network with your 6 main sites being at the top. I can provide you a better design after I see what you have now.

Jason Smith

www.smif101.com

[h.bradley]

I have one EIGRP AS. All sites are interconnected with radio. At each main location, the carrier has an ATM switch. One of the main functions of these ATM switches is to provide a termination for the radio hops that fan out from the main site. The PVC's are actually SVC's, I believe.

Because the carrier has meshed all the main site switches, their entire network has become a bridge more or less.

I have to keep EIGRP running on the main WAN router, and like I said previously, would prefer not to add a wack of static routes to that router. I would be more than willing to disable EIGRP on the field routers though. Could I maybe just use the redistribute connected under EIGRP on the main WAN router and a default static route on the field routers?

[ahbanks]

I think your problem is something like Router1 sends eigrp update to Router2, then the SVC is torn down. subsequent keepalives aren't seen, and Router2 removes the Router1 routes, etc.

Do these routers have ATM interfaces to connect to these ATM switches? Have you configured the 'broadcast' keyword under the subinterface and PVC?

Could the ATM/wireless not be reliably sending multicast traffic?

The redistribute scenario you mentioned would not work. Redistribute just injects routes into an AS from staic or another protocol/AS, but does not distribute those routes to other devices. That job is left to the routing protocol.

Take a look at the 'show ip eigrp neighbor' output, maybe post it here. Looking for the uptimes on the peers.

[h.bradley]

Thanks Ahbanks. The routers are actually ethernet connected. I went ahead and removed eigrp from all of them except the gateway. I added static routes on the gateway to about 60 LAN subnets that are on this network. I then got the carrier to remove all meshing. Wouldn't you know, the problem is still happening. Shortly after I reported this, the carrier found a problem with their radio's. Go figure.