Load balancing EIGRP losing packets when one link fails

Bill19795_2 · ‎07-28-2010

I have two 3750’s configured as a single stack. I am doing some testing now and I have a single 2921 router connected to switch1. I have interface Gig0/0 configured as a /30 and Gig 0/1 as a / 30 on a different subnet. I converted the two ports on the 3750 to layer 3 ports and given them the other IP on the /30 network. I have VLAN 1 on the 3750 acting as the default gateway which is the gateway that my PC is using. On the 3750 and the 2921 I am using EIGRP to exchange routes. I have a loopback interface on the 2921 to test connectivity. When I do a show IP route on the 3750 that is my gateway I see two routes in the routing table for the loopback interface going over both of the 2921 interfaces. I then setup a ping between my PC and the loopback interface. When I shutdown one of the interfaces on the 3750 I loose between 1 and 2 ping packets. I would have thought that with the load balancing in place the second route remains and I would not loose any packets.

gatlin007 · ‎07-28-2010

Keep in mind that when you shutdown an interface a packet on the wire at that time will die in flight. Issuing the 'passive-interface x/x' command under eigrp would be a better test if you don’t want to see any packet loss. The actual forwarding decision with equal cost EIGRP routes is based on CEF by default. The CEF hash will always have your traffic going over a single link based on source/destination IP address. It's not packet by packet load balancing; its flow balancing.

Chris

Mohamed Sobair · ‎07-28-2010

This is because some layer-2 encapsulation protocols doesnt react fastly to a link failure and thus doesnt inform higher protocols about this failure.

So, the route was still in the routing table and not yet removed when you shutdown the interface , hence you saw a packet drop in Eigrp.

I strongly recommend you look at Bidirectional forwarding detection if you seek Milliseconds convergence, it will keep track the link and will inform higher protocol about failures based on specific timers.

Please see the bellow link for more information,

http://www.cisco.com/en/US/technologies/tk648/tk365/tk207/technologies_white_paper0900aecd80243fe7_ps6599_Products_White_Paper.html

HTH

Mohamed