Research On Temperature-Aware Energy Optimization Strategy In Data Center

With the rapid development of computer technology and Internet technology,the scale and number of data centers are growing explosively.At the same time,the huge energy consumption of data center has become an urgent problem,and attracted more and more attention.The energy consumption of air-cooled data center mainly includes two parts:IT equipment and cooling system,which account for almost the same proportion of data center energy consumption.A lot of studies declared that the mismatch between server power distribution and cooling supply in data center leads inefficiency of cooling power utilization.Therefore,Temperature-aware energy optimization Strategy of data center has become a research hotspot in academia and industry.Based on a comprehensive study of the existing energy optimization technology in data center,this paper uses the chip temperature to describe the thermal state of server,and conducts in-depth research on the static and dynamic energy optimization strategies of data center.The main works are listed below:1.We have proposed a chip temperature-aware workload allocation strategy for minimizing the energy consumption in data center.Based on the studying of the heat transfer model of server cluster and the chip temperature-dependent leakage power,a non-linear constrained extremum problem is proposed to reduce the power consumption of data center by optimizing the server workload scheduling and the supply temperature of cooling system.Finally,a simulation experiment is designed to evaluate the performance of the chip temperature-aware workload allocation strategy on energy optimization.2.We have proposed a data center heat transfer modeling method that combined the POD method and cross-interference coefficient,and studied the impact of variable air volume cooling system on the energy optimization perfonnance of temperature-aware workload allocation strategy in data center for the first time.In order to describe the heat recirculation among servers,the POD algorithm is used to reconstruct the cross-interference coefficients for different volume of cooling air supply.On the basis of analysis of the power consumption model of seiner and variable air volume cooling system,the energy optimization strategies for data center are proposed to minimizing the holistic power consumption of cooling system and IT equipment.At last,a simulation experiment is designed to evaluate the influence of variable air volume cooling system on energy consumption of data center.3.In order to cope with the real-time fluctuation of workload,the chip-temperature based model predictive control strategies are proposed to realize the dynamic optimization of energy consumption in data center.A state space model is adopted to describe the thermal dynamic state of server cluster,and the power models of server and cooling system are established based on the dynamic power consumption.Then,two model predictive controllers,energy-first and workload-first,are designed to optimize the workload allocation among servers and air supply temperature of cooling system in real time.Finally,the performance of model predictive controller in data center energy optimization is evaluated by simulation experiments.4.According to the energy optimization problem of heterogeneous data center,the server deployment and workload allocation combined optimization strategy is proposed.Firstly,we have analyzed the particularity of energy consumption optimization in heterogeneous data centers,and a server deployment method oriented to global energy consumption in operation process is proposed.On this basis,the server workload allocation and cooling air supply temperature are optimized by chip temperature-aware workload allocation strategy.To cope with the real-time fluctuation of workload,the cooling air supply temperature is adjusted by time interval.Finally,a simulation experiment is designed to evaluate the energy optimization performance of the combined optimization strategy of server deployment and workload allocation in data center.