The problem with alot of monitoring tools, nagios included, is that if you have 100 servers that you want to monitor, if a switch goes down, you could get 100 alerts about 100 hosts being down. Which is not true. The only thing that is down is a switch, or a cable is unplugged. Setting up host dependancies or even parents is not practical for monitoring 100 servers/hosts since actually, one host does NOT depend on the other. There is no reason why boxA would be needed in order to ping boxB. So parents nor dependancies will do you NO GOOD, unless they truely are parents/dependant. In my nagios setup, since I have the entire network mapped out, the ports, the switches, the port used to connect to serverA and it's interace eth0, it's obvious that parents are valid. ServerA does not have a perent of ServerB, but it does have a parent of ServerA-eth0 which is it's interface card. That ServerA-eth0 host has a parent of the port on the switch it's connected to named SwitchA-port10 and so on.