on each node, the management daemon runs every two minutes and takes the time and uses the cputimer Java method to get the java process cpu net usage. It then compares these readings to the sample from previous run and calculates the presumed server load from the pega java process by calculating ((delta CPU) / (delta time)) / (number of processors claimed by the operating system) * 100% and sends the calculated result to AES
so, lets say that cputimer readings indicate that 130 cpu seconds were used in past 120 seconds and the system reports 4 CPU's (which may be four virtual hyper-threads -- there's no way for us to know). Daemon would assess that this Pega node has used 27% of the system's CPU
Now, it's really up to you to assess the "right" value. That depends on your knowledge of other stuff running on the server(s); the actual hardware architecture (presumably virtualized); expected application load, etc. In PegaCloud / PDC we assume that any server reporting > 70% is "running hot" but that's really based on "gut" more than a specific analysis of the virtualization environment virtual hardware threading model and true capacity.