I've always monitored my servers through the actual interface rather than through iLO. Honestly, I never knew you could see OS resources by polling iLO, but one thing that immediately comes to mind is what happens if the server actually loses it's network connection without taking the interface down (such as a vlan misconfig)? It seems that by monitoring via iLO you would never see that host as down, as the iLO address is up regardless if the OS or main network connection is up.
I also didn't realize you could get all x86 server info you would normally get on server through its snmp agent also through iLO either. If you really can do this I guess I would go for monitoring it through the iLO as it's all out of band management. This way even if the server was down or disconnected from it's regular gig ports (if it's say a DL380 or DL580) then you could still get information it sound like from the iLO card in the server as long as it still has power.
Overall I am of the opinion that monitoring OS resources via the OS is the best idea. That includes the network connection(s).
Reliance on the HP ILO to get the data works in my environment because the sysadmins are pretty anal about putting the ILO drivers in the OS (both Windows and Linux). If the sysadmins didn't do that, there would be no consistent source of data.
I also am considering that the ILO itself should be monitored. You might notice that the ILO network interface didn't show up in the ILO list.
Since the ILO can run standalone and gives the sysadmin remote power / console control of the machine, they would probably appreciate knowing if a system located remotely was not controllable anymore.
- Monitor OS resources via OS
- Monitor ILO resources via ILO