problem with unequal cost load balancing

[ggersch]

Hi,

I am having a problem getting unequal cost load balancing to work. I am using EIGRP and have variance set to 3. I have this mocked up in my home lab to duplicate the problem at work. Here is link to a crude picture of the setup.

http://www.gersch.com/greg/eigrp_problem.jpg

The problem is traffic going from the 10.2.1.0 segment on router RTD to the hosts on the 10.20.222.0 segment. One path is via a T-1 link through router RTA to the segment. The other path is via a 1/2 T-1 (shared with a pbx) though router RTE to the same ethernet segment. I am using a 'bandwidth 768' config line to influence the 1/2 T-1 interface configuration. I have several loopbacks off of other routers for testing.

Here is the output of the 'sho ip eigrp topology' command:

IP-EIGRP Topology Table for AS(101)/ID(192.168.3.1)

Codes: P - Passive, A - Active, U - Update, Q - Query, R - Reply,

r - reply Status, s - sia Status

P 10.1.3.0/24, 1 successors, FD is 2323456

via 192.168.1.1 (2323456/409600), Serial0

via 192.168.3.2 (3998720/409600), Serial1

P 10.2.1.0/24, 1 successors, FD is 281600

via Connected, Ethernet0

P 10.2.5.0/24, 1 successors, FD is 2297856

via 192.168.1.1 (2323456/409600), Serial0

via 192.168.3.2 (3973120/128256), Serial1

P 10.1.4.0/24, 1 successors, FD is 2297856

via 192.168.1.1 (2297856/128256), Serial0

via 192.168.3.2 (3998720/409600), Serial1

P 192.168.1.0/24, 1 successors, FD is 2169856

via Connected, Serial0

P 192.168.3.0/24, 1 successors, FD is 3845120

via Connected, Serial1

P 10.20.222.0/24, 1 successors, FD is 2195456

via 192.168.1.1 (2195456/281600), Serial0

via 192.168.3.2 (3870720/281600), Serial1

Here is the output of the 'sho ip route' command:

10.0.0.0/24 is subnetted, 5 subnets

D 10.1.3.0 [90/2323456] via 192.168.1.1, 01:03:32, Serial0

[90/3998720] via 192.168.3.2, 01:03:32, Serial1

C 10.2.1.0 is directly connected, Ethernet0

D 10.2.5.0 [90/3973120] via 192.168.3.2, 01:37:03, Serial1

[90/2323456] via 192.168.1.1, 01:37:03, Serial0

D 10.1.4.0 [90/2297856] via 192.168.1.1, 01:21:51, Serial0

[90/3998720] via 192.168.3.2, 01:21:51, Serial1

D 10.20.222.0 [90/2195456] via 192.168.1.1, 01:37:04, Serial0

C 192.168.1.0/24 is directly connected, Serial0

C 192.168.3.0/24 is directly connected, Serial1

As you can see from the topology, there are two routes to the 10.20.222.0 ethernet segment, and to all 3 of the loopback segments. But in the routing table, I don't get two paths to the ethernet segment, but I do to the loopbacks. Why do the loopbacks work, but the ethernet segment doesn't?

What the heck is going on? I'm guessing its something about the way eigrp treats multi-access segments, but I can't find anything on it in the documentation, cco, or my books.

Anyone have any ideas?

Thx,

Greg

[thisisshanky]

What IOS version are you running on your router, that you see this behaviour. There are several IOS versions with bugs, where, Unequal cost LB and Variance commands doesnt work properly.

One such bug, was first found in 11.1(3)

Bug id is CSCdi72459 - description of bug is as follows:

EIGRP may not retain the best route from topology into routing table when

variance and metric weights

are configured.

Let us know about the IOS.

[ggersch]

Hi,

The production net is running a mish-mash of IOS, with the 3 in this scenario running 11.1(5), 11.2(9), and 12.0(14). And yes, that might be an issue there.

But, this is also why I mocked it up in the home lab, with the exact same behavior. The routers in the example above are all running 12.1(5)T12

This leads me to believe that there is more to the story. Any other ideas?

Thx,

Greg

[thisisshanky]

Hi,

I simulated the same environment. I am able to get both the routes in my routing table, (route to 10.20.220.0 network) with a variance of 3. Now lets forget what I told about the bug in IOS. Can you paste relevant configs of your all your routers.

Shanky

[ggersch]

Doh! I finally found the problem. One of the production routers, RTE in my example, did not have a 'no auto-summary' line in its eigrp config. I was going over the configs with a fine-tooth comb one more time before posting here when I found it. The odd thing is that the 'sho ip route' on that box did not show the usual signs of being in auto-summary mode. Must be a side effect of the old 11.1 ios code.

As for the EIGRP mechanics of how that was messing up the other router, I have no idea. But as soon as I added the 'no auto-summary' on that leg of the triangle, everything started working as expected.

The home lab test setup was a bit weirder. It took a reload of the RTD router to get it working. I've thrown a ton of things at it since, and I can't get it to break. I'm guessing some config snippet was left in memory and confusing it. (I do lots of changes between reloads.)

Thanks for the thoughts and the support.

Greg