Internal BGP - how many 'neighbor' statements? 2

[Morini]

I've got 3 routers which I'm going to connect to each other using 2 switches, one subnet on each switch. I'll then run internal BGP between them (each router has one or more external BGP peerings). When connecting like this, should I give each router a loopback address and use these addresses in the neighbor statements (making 2 neighbor statements in the config on each router) or should I use the IP addresses of individual interfaces (making 2 neighbor statements per router (ie. 4 in total) in the config)...or does it matter?

 

[KiscoKid]

Doesn't matter which address you establish peering on albeit I prefer to use loopbacks only because that's typically the address people recognise and can usually relate to a particular router.

I suppose which address you peer on depends on your (and your colleagues) familiarity with the addresses in place.

 

[mainstreetexpress]

For running iBGP, I generally use loopback interfaces and let the IGP carry the routing for the loopback interfaces.

In a meshed core (like the one you will run) you are not reliant on single physical connections being up in order for your iBGP to be up, i.e. your IGP will take care of loopback routing

 

[Morini]

Thanks for that - will go with loopback address distributed via OSPF.