I have used EIGRP with the variance command for years. Recently we swapped out our 6513 for 2 Nexus 7K's and lost the variance command. So now I'm stuck running our two MPLS routers to a 3750 which then connects to the 7K's, so I can get the unequal load balancing back in operation.
So because I can't play with it while it's in production I'm using dynamips to simulate the configs. The problem is I'm not able to get a traffic share count ratio that reflects the actual difference in bandwidth.
The two lines have 155 MB and 45 MB respectively, for a ratio of 3.4. Now you can't really send ".4" of a packet but the total packet ratios when dividing the traffic share count should be similar. I'd settle for 3. I was able to get 2.9 in the past but now on the simulator I can't even get that. I'm seeing traffic share counts like 120:89 and 120:71 depending on how I fudge the metrics when I inject the routes from BGP.
I found this link: http://blog.ipexpert.com/2010/05/03/eigrp-unequal-cost-load-balancing/ a stellar article by the way, and this guy is doing calculations to arrive at a 5:1.
BUT when I apply the same formula using specifics from my network and the desired ratio of 3:1 what I get is an AD so high that the route is no longer feasible. I simply cannot make a ratio fo 3:1 using the formula. Has anyone been able to make anything other than a marginal ratio actually work or is this all just theoretical?
I believe I may know what the problem is at least in my lab: for dynamips there is not really any kind of "switch" emulator, not in the sense that it would have all the features of a 6500 or 3700 series.
So I can't set the Bandwidth after calculating it like I would on a physical switch interface. Instead I was trying to set this up by changing the BW metric on the redistribute bgp xxx inside the EIGRP router command set. Of course that's never going to work because as soon as I mess with the redistribution metrics, I've changed the values I was using to calculate with. DOH!
I knew that I had this working in the 'real world' before I replaced my 6500 series router...I just wanted to recreate it in the lab before I tried to reintroduce it into production. But this is a limitation of dynamips. I'm going to have to try this with real equipment to see how the formula is supposed to work.
Right now the 2 mpls routers and the '6513' (simulated by another router) are all on a switch inside the dynampis config. Let me see if by direct connecting them using serial or eithernet ports, I can maybe get to a place where I can have individual interfaces to set the bandwidth command on. That's what I would do in the 'real world' is each router would have it's own physical port on a switch for me to configure at will without impacting the other one...
Yep, I got it now. The problem was the fact that I couldn't really set a bw value on the interface coming into the switch where the routes needed to converge.
I "faked" it by creating a different network for each router to attach to my '6513' (simulated by another router). Did the calcs, put that resulting BW value on my interface and viola! I have a 3 to 1 ratio.
I knew I'd done this and had it working but I also recall now learning that painful lesson about dynamips once before.
*sigh* too much wine since then perhaps?