Power management using dynamic power state transitions and dynamic voltage frequency scaling controls in virtualized server clusters

Reducing power consumption and maintaining user or application performance criteria is an important goal in virtualized server cluster system design. Achieving this multiple objective requirement in a virtualized environment is a challenge. One of the techniques widely explored in the literature to achieve this goal is the dynamic voltage frequency scaling (DVFS) approach. However, power consumption reduction due to DVFS is far less than what can be achieved with the dynamic power management (DPM) control approach to either switch OFF the server or to transition the server to a low power SLEEP state when not processing application requests. In our work, we have formulated a power optimization problem that meets the defined performance criteria for the workload. We have considered application request response time as the performance metric. The adaptive controller is designed to track application performance in the first step, with dynamic control and batch requests to the best server in order to minimize the power consumption as the second step. In this paper, our aim is to reduce the power consumption of a virtualized server cluster by making use of an adaptive hybrid approach that identifies the right server with the best performance for power metric (PPM), DPM with server processor sleep state(s), and DVFS controls. Simulation results show that our hybrid approach using PPM, DPM with sleep state, and DVFS controls is effective, achieves better energy savings, and meets the performance criteria constraint.