Neuro-fuzzy controller in real-time feedback schedulers

Abstract

Traditional scheduling algorithms worked on closed and highly predictable environments. However present day systems need to work in more open and unpredictable environments; such as mobile robots, on-line trading, e-commerce, multimedia that cannot be driven well with traditional open-loop algorithms. A new scheduling paradigm, feedback control scheduling, therefore has been presented recently to fulfil the requirements of such systems. This algorithm defines error terms for schedules, monitors the error, and continuously adjusts the schedule to maintain stable performance. When PID (Proportional-Integral-Derivative) controller is used to control the CPU utilization, one of the problems faced is that when utilization setpoint is closer to 100%, in severely overloaded conditions, systems can have a longer settling time than the analysis based on the linear model since utilization feedback saturates at 100%. To overcome this problem, a neuro-fuzzy controller is designed instead of PID. Simulations showed that settling time with the neuro-fuzzy controller is approximately four times shorter than the one with the PID controller.