We monitor CPU and
memory utilization with cacti (a related topic for cacti monitoring https://forums.ruckuswireless.com/ruckuswireless/topics/anyone_using_cacti )
In the past with a ZD3k we’ve set an alarm threshold for the CPU utilization of 60%, relative to the load we had.
It depends on how many APs you use and the load (in terms of number of clients). In our case (multiple SSIDs, 2 802.1X SSIDs, DVLAN) the CPU of ZD3k spiked at 90%-100% several times with 6k simultaneous clients, 450APs, resulting in ZD unresponsive. These numbers were less than the theoretical limits of 1k APs and 10k clients for ZD3k. Since we changed to a ZD5k, the CPU did not went above 15%.
We also did a http health check for the ZD web interface (script which sends an alarm if the ZD url is unresponsive).