EIGRP Composite Metric

[birenderrana]

By default while calculating the composite metric in EIGRP,  we use k1 and K3 i.e Bandwidth and delay . Same has been prescribed by CISCO as well . What happens when we use other metrics such as k2 , k4 and K5 ? How will it affect the neighbor relationship ?
Please use k values separately while explaining.
Thanks in advance .

[Abzal]

First of all K values mistamtch leads to neigborship failure so it needs to be the same on both ends. Secondly it affects path-selection in that way that it may be impossible doing equal-cost load-balancing because some k values are dynamic.

K1 - it is discouraged to change the value because EIGRP throttles to use 50 percent of the configured bandwidth. Lowering the bandwidth can cause problems like starving EIGRP neighbors from getting packets due to throttling. And may have impact to QoS too.

K2 - measure of load reported by the interface, impacts net-throughput of an interface. The higher the value the lower thoruoghput.

K3 - is cumulative delay.

K4, K5 - reliability metric.  Allows protocol packet loss-based path selection. May have an impact for VoIP, streaming and video traffic because services need reliable path.

Details:

https://www.cisco.com/c/en/us/products/collateral/ios-nx-os-software/enhanced-interior-gateway-routing-protocol-eigrp/whitepaper_C11-720525.html#_Toc443979301

[Cristian Matei]

Hi,

     There are 5K values for EIGRP classic mode and 6K values for EIGRP named mode. These K values can just be set (equal to 1) or unset (equal to 0), and based on this, different metric components are taken into account when computing the metric.

     The value (1 or 0) for each of the 5/6K components, are leaked in the EIGRP HELLO packets, and for adjacency/neighborship to be formed, these values need to match (if one router has K1, K3 and K5 set to 1 and K2,K4 and K6 set to 0, the other router needs the same config). The reason for this is loop prevention; by forcing K1-K6 values to be in sync between neighbors, this guarantees that both neighbors use the same components for metric computation, thus providing loop free routing.

Regards,

Cristian Matei.

[birenderrana]

Hi christian
i understood that if the k values are nor same it will not form neigh and at the same time it used to avoid loop ., but my question is if at both end we have k2,k4 & k5 set to 1 and k1 & k3 set to 0 then what will happen ?

[Giuseppe Larosa]

Hello @birenderrana ,

actually the effective EIGRP implementation takes a snapshot of load over interface and reliability to avoid to create an instable network.

In addition to this you need to consider that the only components that is cumulative in the EiGRP metric is delay. For the bandwidth the lowest bandwidth on the path is considered.

So in real world production networks the default k values are used with K1=K3=1 and K2=K4=K5=0.

Of course in a lab environment you can make tests with different sets of k values but you should realize what I have noted above that load and realibility parameters are static and do not change over time and that without the delay component the metric has no way to increase over multiple router hops or over slower links.

The effective implementation of EIGRP was explained in the forums by Russ White some years ago.

So using the default values of k parameters is recommended.

Hope to help

Giuseppe

[Richard Burts]

I admire the explanation from @Giuseppe Larosa and especially the part about how even though we consider reliability and load to be dynamic values EIGRP does not really treat them as dynamic but uses a snapshot of the values. But I think this does not address the important question in the post to which he responds:

if at both end we have k2,k4 & k5 set to 1 and k1 & k3 set to 0 then what will happen ?

If both ends have consistent K values then the EIGRP neighbor relationship will be formed. And if K values are set as described then both routers will calculate the best path in a very different way. It is likely to reach decisions about what is the best path that are different from decisions made comparing bandwidth and delay. I agree that this might be an interesting exercise in a lab environment, but I would be very hesitant to do this in a live network.

[Giuseppe Larosa · Accepted solution]

Hello @Richard Burts ,

the real issue I see with a choice of metric components like the following one:

>> if at both end we have k2,k4 & k5 set to 1 and k1 & k3 set to 0 then what will happen ?

I agree that EIGRP neighboships will be built if the same set of k values are used.

If load and realibility are static snapshots, there is no metric component that can reflect the whole topology going through five routers or two would be the same or having a very slow link on an alternative path would not be taken in account.

This kind of metric calculation is useless in real world where we want the metric to reflect the actual topology and the speed of each link between routers exactly like OSPF metric does for example.

Hope to help

Giuseppe