A method and system is provided for monitoring the health of processes running on a router. A behavior of a process is monitored and the process is killed if the behavior is abnormal. The behavior may be abnormal if the process is non-responsive, cannot start, or repeatedly crashes. The system may include a timer to measure a predetermined time interval for the process to perform a desired action, a counter to count a number of times the process fails to perform the desired action before the timer expires, and a controller to kill the process when the counter exceeds a maximum number of failures. Alternatively, the timer could measure an amount of uptime, the counter could count the number of times the process crashes, and the controller could kill the process when a crash rate calculated from the number of times the process crashes per the amount of uptime exceeds a maximum crash rate.