Thermal Flow Simulation And Energy Consumption Optimization For Data Centers

In data center design and optimization, the first step is to set up the configuration of the room, racks and cooling equipment. Designers will use computer for modeling and simulation which can help them find the most energy-saving solution. In this thesis, we have designed and implemented a computer-aid design system for data center which provides the ability of 2D and 3D physical modeling, fluid simulation, automatic air conditioner control and energy-saving optimization. Firstly, there exist many limitations on traditional data center modeling software. Moreover, they are costly. Different from them, our modeling tool is simple and practical, and has rich features. Secondly, in fluid simulation, the traditional computational fluid dynamics (CFD) algorithm is time-consuming. We utilize the Euler method to solve the fluid equations. When the grid resolution is large, a multi-grid solver is employed to accelerate the computation. To demonstrate the effectiveness of the simulation, we present a generic algorithm for estimating the parameters used in fluid equations. Thirdly, in energy-saving optimization, current researches focus on the design of the equipment physical size or single parameter optimization. In this thesis, we divide the energy-saving optimization into server power distribution and automatic air conditioner control. For server power distribution, we consider two factors: the distance between servers and cold air inlets and the balanced server outlet temperature, and two weighting parameters are added for each factor. We have designed a fast algorithm for solving the optimization function. Automatic air conditioner control is based on real-time sampling. Moreover, we have designed three server power distributions which allow for changing the server power dynamically. More than 1500 pattern experiments are conducted which shows that different layout parameters will also affect energy consumption. Our system and optimization method will greatly facilitate the design and research of data center.