NBMA is for most network engineers „cloudy“ concept. Lets review little bit  what it is. When we talk about NBMA in cisco world you are most probably refering to Frame relay network. Not because Frame relay is the best technology or newest, but its kind of the foundation technology. To have some overview read the frame relay tutorial at 9tut.com web page.

Frame relay ist second generation technology in packet switched world. Before was x.25 there and after ATM, MPLS.

In Frame relay your connections are defined by PVC (permanent virtual circuits) each one of them having bandwidth limitation known as CIR (committed information rate), which means that provider is eligible to drop your traffic when you consume resources above your CIR and there is no more available BW. Most of the confusion regarding EIGRP when comes to FR is, when you look at hub and spoke topology. You got one HUB connected via one serial link to MORE different devices. In beautiful world without problems you should have only one device facing the serial connection right? Thats why EIGRP behaves so strange in this concept. Then addressing.

Addressing in FR is in the other way confusion for the network engineers. We call them DLCIs (DataLink connection identifiers). There are no MAC addresses at FR world, just DLCIs (numbers). They are locally significant and there is good analogy for them how they work. They work exactly as an airport gateway. You MUST choose the right gateway at your source airport to land at the right destination. You can land at the same gateway, but you dont have to. Important is that you land at the right destination and gateways are locally significant at every airport. And important is to choose from the source the right gateway, you are not searching the gateway of the destination airport when you are going to take of. Similarly routers works with FR. They must choose as a destination the right DLCI allocated to them, to reach the right DESTINATION. Not the DLCI which DESTINATION has.

Frame relay can have different PVC designs:

• HUB and SPOKE – very cheap, but when hub goes down you are fucked up
• Full Mesh – very expensive, but fully redundant and very good for like VoIP. You see this in MPLS world
• Partial Mesh – kind of hybrid, critical sites got redundancy and some „I dont care sites“ doesnt.

How EIGRP handles NBMA? It depends! How many neighbors you got? How your topology looks like. How many subnets you are using, how many PVCs you got…

NBMA stands for NON BROADCAST MULTIACCESS. It means that broadcast along with multicast are forbidden. But EIGRP is using mutlicast 224.0.010 for neighbor establishment and without it we cannot establish the neighborhood! So what cisco did is they tuned they FR and implement thing called „pseudo broadcast“ or „manual neighbors“. You can use keyword „broadcast“ so you allow the broadcast at interfaces. It is just unicast sent to every PVC if you use the keyword for every one of them. Second option is to define the neighbors manually. When you do that you disable multicast for EIGRP, so you must define all the neighbors! When you are going to use point to multipoint topology you must use keyword multipoint after the serial interface command. Default is point-to-point.

One more problem can appear in NBMA networks and that is SPLIT HORIZON. In Split horizon you don’t send routing updates regarding some network towards the interface you received it from. Split horizon is disabled on physical interface but enabled on sub-interface. So when you have setup with sub-interfaces using hub-and-spoke topology, you will not see networks from spokes (branch offices) unless you disable split horizon. You can disable split horizon at interface level.

Also note when you got multipoint topology EIGRP can use by defautl 50 % of the link bandwidth for itself. But this 50% is further split by every neighbor it got. You can increase the bandwidth percentage in interface level configuration. But remember you configure the percentage totally used.