Real-Time Power Modeling And Control In The Power-Aware Dbms

Energy efficiency is an important concern in computer systems from small handheld devices to large scale data centers and supercomputers. High energy consumption can not only affect the availability and reliability of systems, but also may have a serious negative impact on environment; therefore, it is imperative to improve the energy efficiency of systems.Effective power management mechanism is essential to improve energy efficiency. Power profiling and modeling is the foundation of power management. In this thesis, we constructed a power model to predict the system power based on the utilization of resources; and proposed a power controller to regulate the system power. The contributions of the thesis are as follows:1. We proposed a system-level power model which is referred to as software power meter. Here, we improved the prediction accuracy of the power model by incorporating CPU frequency and each core’s utilization as its variables.2. A process-level power model has been put forward by allocating the dynamic system power to the active processes, and thus the software-based power meter can supervise the system power as well as the power of the processes.3. A DVFS-based power regulation scheme is presented. With the help of the power model and the closed-loop feedback control system, we build a power controller that controlls the system power by regulating the processes’power consumption.The experiment results show that the average relative error of the system-level power model is less than12%; and the closed-loop control system is promising to manage the system power in the high power consumption environment, it can improve the energy efficiency by6.53%, with only3.92%performance loss.