Saturday, July 5, 2008

EIGRP - Tuning Convergence Times

Unlike OSPF, EIGRP hello and hold timers (hold timer is like dead timer in ospf) do not have to match between neighbors. It would make sense that if you want fast convergence in one direction you would want it the other direction as well. However, for this example, we are going to tune convergence one way.

The Network:


R1-R2 are connected via frame relay on
R1-R3 =
R2-R3 =

Each router has a loopback advertised into EIGRP. We are testing from R2's loopback to R1's Eigrp hello/hold timers on frame-relay interfaces are 60 and 180 seconds respectively. This means for traffic from R2 to R1 to failover and use R3 it would take 3 minutes to failover!

That would look like this:

R2#ping source repeat 1000

Type escape sequence to abort.
Sending 1000, 100-byte ICMP Echos to, timeout is 2 seconds:
Packet sent with a source address of

You get the idea.

Let's adjust the timers on R1's frame-relay interface, in this case serial 1/2:

R1(config)#int s1/2
R1(config-if)#ip hello-interval eigrp 1 5
R1(config-if)#ip hold-time eigrp 1 15

If we shut down R1's s1/2 interface it should take R2 15 seconds to realize R1 is down and switch over to R3's path to

Let's try it out:

R2#ping source repeat 500

Type escape sequence to abort.
Sending 500, 100-byte ICMP Echos to, timeout is 2 seconds:
Packet sent with a source address of

switch to R1:

R1(config)#int s1/2

Back to R2:

Success rate is 98 percent (493/500), round-trip min/avg/max = 4/32/112 ms

As you can see, this was much faster than the default timers and we only lost 7 pings.


  1. sorry dumb question : what happen if we tune the timing convergence as minim as we can, or how is the minimum value for the time convergence on eigrp


  2. Well if you tune the timers to low, than a few lost packets could cause an unnecessary re-convergence. The lowest values for hello and hold time are 1 second. I don't think that would be wise because 1 packet loss doesn't necessarily mean the router has failed.

  3. question for another topic, but still about EIGRP, what is the big different between load balancing and failover?
    and how we can apply failover scenario on EIGRP

    or is tuning convergence that you posting on category of failover mechanism? (sorry my english is so bad :) )

    thanks for your respond :)

  4. Sorry I did not see your comment sooner. Load balancing is when multiple paths are active and used. Failover is when one path backs up another, only one is active and used at any moment.

    In my example, re-convergence is possible because there is multiple paths. This would be considered failover.


  5. well, i had a bad experience with eigrp simulation on dynagen, well, technically it added a little bit to my knowledge, the eigrp interfaces kept flapping all the time. i had to tune the timers in order to fix this problem. i set the hello interval to 1 sec and the hold timer to 40 sec. well, technically speaking the hold timer is the goal in that case because flapping occures when the hold timer reaches zero. the hello timers minimum value is only to assure that even if lots of hello packets are dropped for any reason, the huge amount of hellos will hopefuly pass atleast 1 packet in the 40 second period.

  6. how would you address the issue with mpls provided metro-ethernet networks that i have seen where a line is "half-up" basically one way communication. so one side is hearing hellos but the other is not, so the offending side will still have the "bad" route installed.

    i thought of BFD with 12.4 but with EIGRP it seems it is only supported on a limited list of hardware.


Note: Only a member of this blog may post a comment.