Performance Control for Nonlinear Hadoop-Mapreduce Computing Systems.

This paper addresses the issue of model establishment and performance control for nonlinear computing systems based on Hadoop-Mapreduce platform. The parameter mapred.map.tasks is selected as the control input data and the CPU usage is selected as the output data. By using least-squares principle, a Hammerstein model for this relationship is identified and validated. The control law consists of a nonlinear compensator and a PID feedback. By testing in Simulink, the Hammerstein model is verified appropriate and the designed nonlinear control drives the CPU usage to achieve the target value asymptotically. [ABSTRACT FROM AUTHOR]