The purpose of EIGRP Split Horizon

[Sam-CCNP]

Hi anyone,

Quick question, but am I right in assuming that the purpose of Split Horizon in EIGRP is to suppress unnecessary updates rather than to prevent routing loops?

DUAL (and the FC) seems to take care of routing loops, so I'm thinking that is its only point.

Sam

[Sam-CCNP]

Hi @David Ruess ,

FYI, Brian McFadden's response as to why split horizon is needed in EIGRP....

"There are no cases where disabling split-horizon in EIGRP will cause a routing loop.  This is due to the feasibility condtion (FC), as previously mentioned in the thread.  The reason why is the following logic:

FC says that if your metric to the destination is lower than mine, you are a loop free path. (my metric is X, your metric is < X to become a feasible successor).

If I advertise a prefix to you with metric X, and because of split horizon being disabled, you advertise the prefix back to me, the metric will always be greater than X.  This causes the feasibility condition to fail, and the route to be rejected.

 The only reason split horizon is used in EIGRP is to prevent unnecessary duplicate routes in the topology table, which eats up control plane cycles when the router needs to run DUAL in order to converge.  Count to infinity can't happen in EIGRP like it can in RIP, due to the feasiblity condition and the QUERY/REPLY process when a convergence event occurs."

why split horizon needed in EIGRP? (cisco.com)

That makes perfect sense to me.

Sam

[David Ruess]

Great. As long as you understand it that's what matters. 

---

If I advertise a prefix to you with metric X, and because of split horizon being disabled, you advertise the prefix back to me, the metric will always be greater than X.  This causes the feasibility condition to fail, and the route to be rejected.

 From the statement you can also say If I advertise a prefix to you with metric X (even if SH is disabled) and you are using me as your successor I will advertise that metric back to you as infinity. However, If I learn that same route from somewhere else as well and that becomes my sucessor then I would advertise that route to you with a greater (non-infinity) metric.

EIGRP (as are all routing protocols) are built for all kinds of topologies and prevent loops and provide best path routing for every single one of them. Some functions like Split Horizon, Poison Revers, DUAL, Feasibility condition all work synergistically to have the same outcome of being loop free. Some features work together while others work separate and depending on the topology each feature can have multiple uses in preventing loops.

-David

[Elliot Dierksen]

I haven't had to think about this in quite some time, so I could be a bit rusty. My understanding of the premise is that if there were multiple neighbors on an interface, that interface is ASSUMED to be broadcast capable. That means all the other neighbors SHOULD hear the advertisement on their own, so there is no need to advertise it back out the same interface. I can only think of two use cases where disabling split horizon was beneficial/desirable. The main one was for  DMVPN hub. The other was if you were using frame relay with 'frame-relay map ip' statements instead of using point to point sub-interfaces. You don't see too much frame relay any more, but I am firmly in point to point sub-interface camp.