4M-Switch: Multi-mode-multi-model supervisory control framework for performance differentiation in virtual machine environments

Abstract

When resources are shared in a virtual machine environment, providing different performance levels to different customer applications is a challenging task. In order to sustain stability, the control solution not only has to take into account the time-based dynamics, but also has to adapt to various operating modes. This paper proposes the 4M-Switch supervisory control system design framework, which takes into account the possible operating modes and dimension changes of the VM environment at design time and then adapts the control solution to achieve required management goals when changes occur at runtime. 4M-Switch utilizes a piece-wise linear modeling approach to present the behavior of the system using multiple models and simple switching logic to change the controller parameters to mitigate the effects of nonlinearities. The experiment results conducted under a range of conditions show that 4M-Switch approach effectively adapts the control solution and provides significantly more stable performance differentiation compared to the existing approaches.