A service's CPU load percentage is calculated once during each monitoring interval as specified by the -f attribute. 

If a service's CPU load exceeds the threshold specified by the -c option, the service is killed. However, the threshold can be ignored for a specific number of monitoring intervals by specifying the -i option. In this case, the service will only be killed if it exceeds the threshold level during every separate monitoring interval.

For example, the options -f 30 -c 50 -i 5 will monitor a service every 30 seconds, and if the service's CPU usage exceeds 50% during each of the five intervals (i.e., 30 x 5 = 150 seconds), then it will be killed. 

Note: The -i option also works in conjunction with -m option.