# How2pass.com Forums

Full Version: eigrp variance
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
i have a question on the eigrp variance question (it's easy to find the question- just go to the eigrp section and skim through the questions until you see one that has answers:
-3
-4
-1
-2
-5

the lowest cost path is 10 + 10 + 10 = 30
multiply that by variance 2 equals 60

i see four routes that have a variance of 60 or less
AFGH
AFEH
I personally think the answer is 2 paths... ADEH being the best path... AFEH being the alternate path.

You said 4 paths... but this is wrong because the variance rule states that that the alternate paths must be LESS THAN  (variance X local best metric)  The other 2 paths are EQUAL TO 60, which is the variance of 2 multiplied by the local best metric 30.

"The metric of the entire path (the FD of the alternative route) must be lower than the
variance multiplied by the local best metric (the current FD). In other words, the met-
ric for the entire alternate path must be within the variance." -CiscoPress book

Do you agree?
BTW, the question ID is RTE029 so that you can search for the question.
I agree with you.  I also 2 for the answer.
So, how do we get H2P to correct RTE029?

You said 4 paths... but this is wrong because the variance rule states that that the alternate paths must be LESS THAN  (variance X local best metric)  The other 2 paths are EQUAL TO 60, which is the variance of 2 multiplied by the local best metric 30.

"The metric of the entire path (the FD of the alternative route) must be lower than the
variance multiplied by the local best metric (the current FD). In other words, the met-
ric for the entire alternate path must be within the variance." -CiscoPress book

Do you agree?

I disagree. According to Cisco, the alternate paths' feasible distance must be less than or equal to the (feasible distance * variance) value of the lowest FD path to be considered for load balancing. However, in order for a route to be considered for load balancing, it must also be a feasible successor. Therefore even though there are 4 paths with feasible distances that would qualify them as load balancing candidates when multiplied by the variance value, only 3 of them have advertised distances that qualify them as feasible successors, so those are the only paths that will be used.

I know this thread is old, but this should clear up a few things for those who may stumble across it in the future.
I think Mick's answer of 4 is correct, but the reasoning behind it is flawed in a way that might be helpful to address. One thing to remember about routing decisions is that a router doesn't choose the complete path it will send a packet along; it can only choose the next hop.

In light of that, ADEBCH would not be a path here. Let's say a packet gets forwarded from A to D and then from D to E. The focus is now on where E will send the packet. E has a route to H with a metric of 10. Its route through B has a metric of 40. With a variance of 2, E would not send the packet to B. It would forward it directly to H.

AFGH would also not be a path. Let's say a packet gets forwarded from A to F. The best path from there goes through E, and even with the variance command, it would not load balance between E and G because the advertised distance from G is not less than the distance through E. It doesn't meet the feasibility condition.

Now for a few words on what the answer may be.

A lot hinges on how the variance command works. We can all agree that to qualify for unequal load balancing, a route needs to meet the feasibility condition, and it's well established that the FC is a solid less than.

But Cisco documents I've looked through are vague on the exact variance algorithm. Testing on actual equipment shows that if the variance-adjusted metric is equal to the successor's metric, load balancing does occur.

Unfortunately, the answer to this question depends completely on how the question's author understands operation of the variance command.

I set up a rather elaborate test of this scenario which I documented in a different thread (search for the QID and you'll find it), and I found that traffic gets load balanced across 4 paths: