This decides whether the device, on each authentication attempt, always starts with the highest-priority configured server, which adds latency if that server is actually down, or remembers and prefers the last server that successfully responded, which speeds up subsequent logins during a primary-server outage but might delay noticing the primary has come back. For example, setting this to 'cached' during an extended primary-server outage avoids every single login attempt re-trying the dead primary server first before falling through to the working backup.