I just completed setting up new NPS servers to serve our network. We have three domains with two servers each. Theses are selected through two Aerohive AP230s running as proxy servers. During my testing I would use the Radius tools in Hivemanager to ensure I could connect and that the fail over was working. When I simulated a failure of the active NPS server it took several attempts before it would switch. Now there is a difference in that the tool does a single request (I believe) or very few in comparison to an actual request which send multiple requests to the server. So in real life the failure will be quicker but not instant. And will be different if you are using a Proxy or going direct.
