Behavior of EIGRP and ISIS routing not as expected

[Can_networking]

I three 2811 routers setup as follows:

Traffic enter R1 via fast ethernet port 0

R1 is connected to R3 via a 115200 b/s serial link

R1 is connected to R2 via fast ethernet connection 

R2 is connected to R3 via fast ethernet connection 

Traffic exits R3 from fast ethernet port 1

In summary

R1-R3 is a short path with low bandwidth

and R1-R2-R3 is a long path with high bandwidth

I noticed the following behavior with the following routing protocols

1) RIPv2, path R1-R3 is selected, unless I fail the serial link then path R1-R2-R3 is selected. This is expected since it is a distance based protocol

2) OSPF, path R1-R2-R3 is selected, unless I fail the etherlink R1-R2 then path R1-R3 is selected. This is expected since it is a state based protocol

3) BGP behaves like RIPv2 for the same test. This is expected since it wants to minimize transit costs.

4) However with EIGRP and ISIS routing I get contrary results to what I expected.

EIGRP behaves like OSPF, I would have expected it to behave like RIPV2 (a distance based protocol) ???

ISIS behaves like RIPv2 , I would have expected it to behave like OSPF (a state based protocol) ???

[Peter Paluch · Accepted solution]

Hello,

The path choice in routing protocols is determined by the nature of metrics they are using - not by the underlying type of the routing protocol (whether it is distance-vector or link-state).

EIGRP takes the minimum bandwidth and total delay into the computation of the resulting metric. The higher the minimum bandwidth is, the lower - and better - the metric. So EIGRP basically prefers the same quality that OSPF prefers - the bandwidth of the route toward the destination. That makes EIGRP do the same choice as OSPF in your network.

With IS-IS, there is no automatic recomputation of interface bandwidth into IS-IS costs, contrary to OSPF. All IS-IS interfaces have the default cost of 10, regardless of their speed. Running IS-IS in this network essentially makes it prefer paths with the least total numbers of links to the destination, and that is why IS-IS appeared to make the same choice as RIP.

Once again, the way a routing protocol makes its best path selection is based on the nature of the metrics it is using. The underlying mechanism of a routing protocol affects the overall information about the network that the routing protocol can collect. In distance-vector protocols, routers send messages that contain arrays (in software engineering, an array = a vector) of {Network, Distance} entries toward known destination networks. In link-state protocols, routers send messages that contain information about each particular router and its links to its adjacent objects (other routers or network). In distance-vector protocols, none of the routers knows how the entire network precisely looks like. In link-state protocols, every single router knows precisely how the entire network looks like.

Feel welcome to ask further!

Best regards,
Peter

[Can_networking]

Thank you Peter for your excellent reply.

I did some further research and found a useful command for EIGRP

show ip eigrp topology 192.168.1.0 255.255.255.0

It clear shows the BW, distance (in terms of delay), load  and reliability metrics used for the routing decision.

[Can_networking]

My understanding from the second part of your answer is that the interface costs of the IS-IS routers must be user provisioned and then the network can behave like an OSPF network based on link state.

[Peter Paluch · Accepted solution]

Hello,

the interface costs of the IS-IS routers must be user provisioned and then the network can behave like an OSPF network based on link state.

If you want IS-IS to differentiate between faster and slower links, then you are correct - you have to manually configure interface costs so that faster interfaces have a lower IS-IS cost, and vice versa. However, regardless of the cost setting, IS-IS is a link-state protocol just as OSPF, and it always makes its decisions based on link-state rules. Just with all interfaces left at their default cost of 10, IS-IS treats every interface as equal.

Best regards,
Peter

[Can_networking]

Thanks Peter,

It all makes sense now.

[Peter Paluch]

Hi,

You are welcome :)

Best regards,
Peter