EIGRP algorithm for unequal bandwitdh

[hs08]

Hello,

If we configure the EIGRP to load balance across 2 line with different speed let say link-1 is 20Mbps and Link-2 is 50Mbps, how actually EIGRP will do the load balance? Will EIGRP send to Link-2 more traffic than Link-1 since Link-2 have higher bandwidth?

[David Ruess]

You are partially correct. It will proportionately load balance the traffic. However in your example its not best to use the BW as a metric as EIGRP will only use the minimum BW along the path in its metric calculation. So just because one exit link has a higher/lower BW than the other doesn't mean it will load balance like that. EIGRP will load balance proportional to its Computed Distance metric. So if one route has a metric of 1000 while the other link has a metric of 500 it will send twice as much traffic over the link with the metric of 1000 than the link with a metric of 500. 

You can verify this with the show ip route <ip> command. In there is a section called traffic share count. If all paths are equal it should be 1. It its unequal you will see different numbers.

Hope that helps

-David

[hs08]

HI @David Ruess 

If change the bandwidth is not best to use, what should we do to make EIGRP load balance as proportional on different link speed?

[David Ruess]

You can either adjust the interface delay along the path or you can add to the metric with an offset-list. This calculation is made per router so if you want to load balance across a path you may need to make adjustments on all routers along the path.

[hs08]

Let say the link speed is 10mbps and 20mbps, so we set delay twice for interface with low speed? Example for link 10mbps we set 100 and delay for link 20mbps se to 50? With different delay value eigrp will not choose slow link as backup line?

[David Ruess]

Its not proportionate like that. The EIGRP metric is based on 2 things: CUMULATIVE BW and TOTAL delay. So of all the links the path goes across it will see if its lower BW than what it already has in its update and for every router hop it will add the incoming interfaces delay all together in order to receive its metrics value.

You see how the 10Mbps link is almost double the metric of the 20 Mbps link. If variance is configured and the higher cost metric meets the feasibility condition then it will likely forward traffic in a 2:1 flow.

[hs08]

so, 20mbps metric is 129280 and 10mbps metric is 258560 this mean link 10mbps will be as backup. Am i right?

[David Ruess]

That's if the minimum BW is those numbers with those delay values only as pictured above. No other links. I was just giving an example. And as I said it will be a feasible successor if it meets the feasibility condition of the route. 

If EIGRP is only using the 20 Mbps link and it fails then by default it will have to use the 10 Mbps link regardless of the metric.

[Joseph W. Doherty]

David is well addressing how to have EIGRP "sees" the different costs, and how you might adjust.

I would like to mention . . .

First, by default, EIGRP does not unequal cost load balance.  You need to adjust the EIGRP variance to allow that, and that adjustment determines how much a difference will be considered to use unequal LB across paths.

Second, when unequal is done, the LB is done, proportionally, by flow count.  For your example, if you get the distance metric across the two path in the same 2.5:1 ratio, that's how flows will be directed to the two paths.  This would be pretty ideal if all flows were using the same bandwidth (e.g. VoIP G711 Codec), but as many flows have variable flow rates, new flows might be directed to a path that's saturated while another might be barely being used.

Lastly (I recall?), one of our early Hall-of-Fame members ( @Edison Ortiz ), very strongly recommended, years ago, not to use EIGRP variance because it caused a huge performance hit.  That may no longer be true, either on L3 switches, or software based routers using CEF.

[Richard Burts]

It seems to me that the original poster is focusing on balancing based on the local links. The point that David and Joseph are making is that EIGRP looks along the entire path to make its decision about load balancing. If the destination were a single hop away then EIGRP would do exactly what the OP wants. But if the destination is multiple hops away then the bandwidth of the local links may not be the controlling element in determining the load balance ratio.

[Richard Burts]

And as I think about it it seems clear that when the OP describes paths with 20 mbps and 10 mbps and wants more traffic on the 20 path it makes sense if the destination is directly reached by those paths. But what if the path with 20 mbps connects to a device and the outbound path from that device is 5 mbps but the next path for the 10 mbps path is another 10 mbps do you still want to favor the 20 mbps path?