A measurement-based power consumption model of a server by considering inlet air temperature

Abstract

The server power is critical not only for IT equipment power consumption, but also for non-IT power consumption in order to provide a suitable environment for IT equipment. The server power model is essential for thermal environment and energy efficiency. Due to the complication of the effects on power consumption of servers and few power models considering the inlet air temperature, this study firstly built an experimental setup to effectively facilitate the factor analysis. It then evaluated the effects of inlet air temperature and CPU utilization on server power consumption. During the experiment, the inlet air temperature was adjusted between 20 °C and 35 °C in 5 °C increments. Simultaneously, the SPECpower_ssj2008 was used to change the workload operations per second from 100% to idle, with a 10% interval. These measurements were also used in the server's various CPU power managements. When the inlet air temperature increases from 20 °C to 35 °C, the highest power increase was 17.60 W for Performance operating mode. Furthermore, in the DAPC and OS operating modes, the server power consumption has a piecewise linear relationship with its CPU temperature, and the function fits well when the CPU temperature is lower than 80 °C. For Performance and Workstation operating modes, the server power consumption is proportional to the temperature difference. The measurement-based correlation method employed by this paper can be used to investigate the power consumption model of other type or configured servers. If so, the adaptability of the model can be improved.