EIGRP Split Horizon seems to be not working on virtual platforms

[NetworkingGeek1]

Hello community, 

I was checking how EIGRP is working on ISRv platform and noticed some strange behavior. Looks like Split Horizon rule doesn't work there. I configured EIGRP adjacencies between two routers: R1 & R2. Let's say there is

network 192.168.1.0 /24

between them. I configured another interface on R1 with

ip address 172.16.1.1 /24

and included it in EIGRP process and this is what I see in Wireshark captures:

R1 sends update about

172.16.1.0 /24 to 224.0.0.10

, R2 sends hello ack to R1 as unicast packet

Then, R2 sends update about

172.16.1.0 /24 to 224.0.0.10

over the same interface in which it received advertisement about

172.16.1.0 /24

, then R1 sends hello ack to R1 as unicast packet. -> To me it's a clear violation of Split Horizon rule - "The split horizon rule states that routes will not be advertised back out an interface in which they were received on."

Could you please clarify, what is this? Is it something I don't understand or might be some ISRv bug?

Thank you.

[David Ruess · Accepted solution]

Hello,

EIGRP utilizes Split Horizon and

Poison Reverse

slightly different than how you may interpret it. If you just have those 2 devices as neighbors then what you're seeing is that R2 is is advertising that same route back to R1 with an infinite metric (as in

Poison Reverse

). Here is a snippet from the RFC that clarifies the process a bit.

It could be a bit more clear with the line: 

EIGRP
   implements this to mean, "if you have a successor route to a
   destination, never advertise the route out the interface on which it
   was learned".

 In terms that it wont advertise it as "reachable" out that interface. Instead we advertise it with an infinite metric which essentially means I'm using you to get to it so don't use me to get to it. 

Hope that helps

-David

[Giuseppe Larosa · Accepted solution]

Hello @NetworkingGeek1 ,

poison reverse

in action here likely the second router will reply with an update back with infinite composite metric to tell it is not an alternate path for the prefix.

You can consider

poison reverse

as a re-inforcement of the the split horizon rule instead of not sending out updates for the prefixes learned on the interface the other router sends back updates that state infinite metric so that if the original router has to find out a new path for the prefix it already knows the neighbor is not a feasible successor.

Hope to help

Giuseppe

[NetworkingGeek1]

Hello @David Ruess  and @Giuseppe Larosa  Thank you so much. I checked it again and indeed R2 sends back routes with infinity delay, which, if I understand it correctly, means infinity metric.

[David Ruess]

You would be correct. Glad we were able to help.