DVFS And Thermal-aware Task Scheduling For Data Centers

With the fast development of cloud computing and data centers, the data centers are becoming increasingly popular world wide. The annual combined electricity consumption of China's data centers accounts for over 2% of China's total electricity use, almost the same as the electricity produced by the entire Three Gorge Dam. Reducing the energy consumption of data centers and making them greener have been the major research.The majority of energy inefficiency in data center is attributed to idle running servers and to heat re-circulation, respectively. In this paper, to solve these issues, we propose a task scheduling algorithm TSTD based on the DVFS technique and the thermal-aware methodology to minimize the total energy costs of data centers. The TSTD can reduce total energy consumption considerably including computing energy and cooling energy. It can resolve the idle running servers issue by adjusting the frequency and voltage of servers which greatly reduce the computing energy and cooling energy when the data center's utilization is low. Since the heat re-circulation is determined by the data center layout and the power distribution. The TSTD can provide a more reasonable power distribution which leads to less heat re-circulation.We compare the proposed TSTD algorithm with existing approaches such as the FCFS scheduling algorithm, the FCFS algorithm with power management(FCFS-DVFS) and the thermal-aware scheduling algorithm SCINT. The simulation results based on the data from real data centers show that: comparing with the FCFS scheduling algorithm, the FCFS-DVFS scheduling algorithm can save 3.7%-42% total energy, the SCINT can only save 1.6%-5.3% cooling energy, and the TSTD scheduling algorithm can save around15.2%-42.7% total energy.