Cooling control restraining effects due to ICT equipment utilization of disturbance based on model predictive control for modular data center

Abstract

This paper proposes a cooling control method that reduces spikes in CPU temperature that occur due to fluctuation in the utilization of information communication technology (ICT) equipment based on model predictive control (MPC) for a modular datacenter. To cope with the fluctuations, the proposed method not only considers the server power consumption using a prediction model, but also switches between an MPC controller and a necessary-air-volume controller based on the rate at which server power consumption rises. The MPC controller controls the CPU temperature in order to do three things simultaneously: avoid throttling the operation of the CPU, reduce as much as possible the power consumed by data center cooling fans, and adjust for the effects of fluctuations. The necessary-air-volume controller calculates the command value of the revolution speed of the cooling fans to supply the air volume required during maximum CPU utilization. The results of our control simulation show that the proposed control method can drastically reduce spikes in CPU temperature. The proposed method provided energy savings of more than 37.6% compared to the conventional control method under conditions where the CPU utilization is 80% and the fresh air temperature is 20°C.