Managing performance and resources in software systems using nonlinear predictive control

Management of quality of service performance and resources in a shared resource environment is vital to many business domains to achieve business objectives. These management systems provide agreed levels of quality of service to their clients while allocating limited available resources among them. It is well known that the behavior of such software systems illustrate nonlinear characteristics, imposing difficulties to model and control the system. This paper proposes a nonlinear model predictive control technique for managing the performance and resources in such a shared resource environment. In particular, a block-oriented Wiener model is utilized to represent the software system as a multi-input and multi-output model in series with static nonlinear components at the outputs. Then a predictive control system is designed by compensating the estimated nonlinearities with their inverse. The simulation results show that the proposed nonlinear model predictive control mechanism has significantly improved the performance and resource management at runtime over the linear predictive control counterpart.