I am looking for some clarification or explanation on how EIGRP load balancing works.
I have a pair of 6500 VSS switches in data center 1 connected to a pair of Nexus 5672's in data center 2. The switches are connected via 4 1GB DWDM circuits. The interfaces on each side are layer 3 routed ports and I am running EIGRP between the two locations. I have two circuits on 6500 A going to 5672 A and two from B to B. I am trying to figure out how eigrp load balancing works between the two locations - which path(s) a source and destination is taking. Both sides are using the default load balancing for the given hardware. I thought the load balancing was just source-destination however it seems like I am wrong about that (at least on the 6500 side)
6500 side
If I do a sho ip cef exact-route x.x.x.x x.x.x.x it shows my a route out of a single circuit. However when I run a replication between the two data centers I see traffic for this source and destination taking two circuits (this was confirmed with netflow). Why the discrepancy and why only two of the circuits rather than all 4?
I also see when I do a sho ip route x.x.x.x for that same destination that the two circuits being used have a load of 1/255 and an * while the two not used have a different load. Does this load influence the load balancing? Where does it get the load? The composite metric for each route is the same.
Does dCEF rather than CEF cause the EIGRP load balancing to work differently?
Nexus side
I haven't really been able to find much on how load balancing works and how to determine a path, etc like on the 6500 since the Nexus does not use CEF. I do not have a difference in the routes on the Nexus like I do on the 6500.
Any help would be appreciated. I am troubleshooting some data replication issues and trying to narrow down if a better load balancing method is needed or possibly bonding the circuits is needed.
Kevin
