Predictive Control for Relative Performance Management

This paper examines the implementation of Model Predictive Control (MPC) for relative performance management in a shared resources system. Finite Control Set MPC as a model-based control is integrated in a scheme of differentiated control for a multi classes virtualized software system. A dynamic model is estimated in block-oriented form with nonlinear compensation feedback system. The performance objective is to maintain response time on the reference levels or subject to the degree of priority between the user classes. Experiments are conducted in a two-classes of virtualized software system for several performance differentiation scenarios. The performance of Integral FCS-MPC is evaluated in comparison with Proportional Integral (PI) control. The results show that predictive control framework provides significant improvement in predictability and disturbance rejection procedure. Therefore, Integral FCS-MPC outperforms PI controller in terms of maintaining the stability of relative performance objectives.