Crazy idea: Identify if system is idle when it's a black box

I need to be able to tell if a series of servers have anyone active on them, and if not then to automatically shut them down (turn off the VM). It's not a trivial task, because I have 1000+ server instances that include an assortment of OSes (Win, Unix, Linux) and many different type of configurations. This makes installing an uptime agent on the boxes non-trivial. Also, because the users are admins, I can't really ensure that tools I install will not be tampered with.

So my idea is to treat each server as a black box and use stats from outside of the server to decide if there's activity:

  • Monitor all servers for disk and CPU activity.
  • If disk writes and CPU activity fall to zero for 1 hr then assume system is idle and shut it down.

I don't care to turn off more machines than are truely idle, if I have something like 90% accuracy. Would the above black box work or be unreliable? What black box metrics would be more appropriate?


Just tell people that if they mess with the scripts, you won't see writes or CPU activity, and it'll kill off the instance :)