EIGRP metrics across L2 via SVI

[Jason Vanlandingham]

If you establish EIGRP neighbor relationships via SVI interfaces across an L2 link, does the metric calculation look at the physical link the same way?

We'll have three devices connected to each other that will require some VLANs to exist on all three, and some unique space to be routed between them. The devices will be connected daisy-chain style by 10G fiber, with a dual 1G link from 1 to 3 for redundancy. Our thought is to have a "transit VLAN" where each device will have its own address in an SVI, then establish EIGRP through those SVI interfaces, but we want to be sure the 10G links look better than the 2x1G.

And is there any tricky config we need to do in EIGRP so that having three neighbors in one segment doesn't cause problems?

[Rich Uline]

Jason,

By default, EIGRP will use only the bandwidth and delay from the Layer-3 interface to determine best path. If you issue the 'sh int vlan [vlan #]' command, it will list the badwidth and delay of the SVI. You can influence these values by using the interface commands 'bandwidth [# in kbps]' and 'delay [# in tens-of-microseconds].

However, if I understand your topology correctly then your initial question is moot. You have two routers communicating via a Layer-2 connection through a third device at 10 Gbps, as well as a backup 2 Gbps etherchannel directly connecting the two routers? This would create a Layer-2 loop which spanning tree would block. I would be inclined to set the middle device in the 10 Gbps sandwich as the root bridge to be sure STP was behaving as expected. Otherwise just check to make sure that the 2 Gbps etherchannel is the link being blocked, and not the 10 Gbps one.

[Georg Pauwen]

Hello,

I am not really sure I understand what your logical setup will look like, but with regard to EIGRP and SVIs, keep in mind that EIGRP by default sees the SVI as a 1Gigabit link when it comes to bandwidth.

[David P.]

Hi,

We're using EIGRP over SVI's and L2 port-channels (2x1G) here, I did change the bandwidth on the VLAN interface to 2000000, but the "minimum bandwidth" reported to the neighbor is still 1000000, as seen in the show ip eigrp topology <prefix> output...

I did the config and the checks on both sides of the neighborship. Show interface vlan X reports 2000000 as configured.

Is there something I'm missing here ?

(Platform is C3850 w/ IOS 16.06.06)

Kind regards,

David

[p.lan]

David, did you ever solve this? we're running into the same issue.

[David P.]

Hi Plannue,

Unfortunately we didn't, to this day the metrics are still the same for every prefix, whatever the configured bandwidth is on the SVI....

Kind regards,

David

[Giuseppe Larosa]

Hello @David P. ,

in EIGRP the bandwidth component of the metric is the minimum value on the path to the prefix.

For this reason changing the bandwidth on the SVI used to peer in EIGRP with the other device may be not enough as only the minimum value is reported in the update in its own separate field.

For example if you are advertising Client Vlans that have their own SVIs you should increase the bandwidth value to 2G on each of them (because the default is 1G)  to see the reported bandwidth value updated to the other EIGRP neighbor.

Hope to help

Giuseppe

[p.lan]

Hi Giuseppe,

I did exactly this. Topology contains several point to point VLANs, with one being trunked across between L3 switches in the network. In a /30 P2P network, I set both SVIs to bandwidth 10g (the physical connection between switches is 10G), but the EIGRP values for the network and for each peer shows bandwidth of 1G.

Any other ideas?

[Giuseppe Larosa]

Hello @p.lan ,

as suggested by @Deepak Kumar you should consider moving to named mode EIGRP that allows to use EIGRP wide metric.

EIGRP wide metric is 64 bit and allows to accomodate bandwdth values of 10Gbps or more.

Standard EIGRP metric is 32 bit wide and has limitations for the higher speed links.

In addition to this as I have explained in my previous post it is not enough to manipulate the bandwidth of the "peering" SVI, but also the one of every client facing VLAN SVI, because only the lowest metric value on the path is advertised in the metric component.

This is also why to create a less preferred EIGRP path is better to manipulate delay that is cumulative rather then bandwidth that can lead to non expected results.

Hope to help

Giuseppe

[David P.]

Hello Giuseppe,

Thank you for this explanation !

In my case there is only 1 SVI on each side of the link, and the bandwidth of both has been changed to 2G, without any effect on the EIGRP metric...

Edit: I just realized that you are suggesting to modify the bandwidth of all "LAN-side" SVI's for those prefixes, this is interesting thank you ! I will try this out when I get the chance !

Kind regards,


David

[Deepak Kumar]

Hi,

As everyone has mentioned that it will take an SVI interface parameters. here, STP  and EIGRP wide metrics also need to consider.

 I think for a better solution, can you share a network diagram and current state of interfaces?