In our environment we have two pollers, a separate database server, and an additional web server. We dual home all these servers to separate switches, and use NIC teaming in an active/passive mode. This provides us fault tolerance at layer-2, so that if the primary switch that our NPM systems are connected to goes down, they immediately start using the NIC attached to the secondary switch those monitoring nodes are attached to. There's no special configuration you need to do with NPM, it's all at the OS layer with the NIC drivers. As previously stated this protects us from a layer-2 failure of a switch; it doesn't help if the layer-3 routing for that VLAN segment goes down, but as long as you're running something like VRRP or HSRP, you're covered there too and your monitoring system should not go down or be unable to alert you of failures.