Control-theoretical stress-testing for cross-platform compliance

Abstract

Modern systems are required more and more to be able to provide performance reliability qualities for not just one system but cross platform as well. Stress testing applications allow for performance testing of a system, however, when applied across different platforms the performance of the application is not a one to one comparison. This thesis introduces the idea of applying control theory to provide consistent resource percentage utilisation when stress testing cross platform. Proportional Integral Derivative (PID) control, Linear Quadratic Regulator (LQR) control and a PID Assisted LQR control were utilised as three control system in this thesis. While the PID controller provided the best results across multiple platforms, its inability to control all the variables in the system along with multiple different state objectives (CPU and memory) leads to the need for a system model approach. LQR controllers provide this model based approach but at the cost of being unable to adapt across different platforms and to unpredictability in the system. Finally, a combined approach PID Assisted LQR control. PID Assisted LQR control gave the best of both PID and LQR control approach and had improvements in the case of where the system was most unpredictable but lacked any significant improvements in areas where the model gave an accurate representation of the system.