This paper deals with data-driven predictive control for relative performance management in virtualized software system. The system dynamics are characterized in Hammerstein-Wiener structure to capture nonlinear and linear characteristics. The proposed control approach is the implementation of Subspace-based Predictive Control with the integration of nonlinear compensation. The compensator functions are inverse static input and output nonlinearity models from the Hammerstein-Wiener system identification. The subspace predictors are formulated from the linear model input and output of Wiener block. The experimental results from three scenarios of performance objectives show the reliability of Subspace-based Predictive Control to manage the virtualized software system.