On 11/Nov/16 08:22, sthaug@nethelp.no wrote:
We have a similar use case, and we run BGP on Quagga. Works great. Haven't seen a need for either IS-IS or OSPF on Quagga yet.
Two reasons for us: * IGP metrics in the IGP will determine latency-based decisions. I know BGP can infer the IGP metric, but we are just avoiding situations where this could be non-deterministic due to other BGP-things. * BGP occurs at a much higher layer in the network stack. We run the Anycast servers in the IGP domain because that is at a much more basic layer. If BGP fails, we don't want to have problems logging into our routers because it took TACACS+ with it. There have been times when BGP has run into issues but the IGP has remained alive. Via a jump host (and OoB, of course), we were able to maintain connectivity to the network to fix it because those servers are in the IGP domain. Mark.