Data Center Job Scheduling Algorithm Based on Temperature Prediction

Abstract

Improved energy efficiency of data center is in hotspot. Prevailing data center energy-conservation measures are intended for computing devices, but ignoring the potential energy savings from cooling equipment whose energy consumption accounts for about 40% of the total energy consumption of data center. In addition cooling equipment is often set to an excessively low temperature to ensure the thermal safety of the data center, resulting in energy waste. In this paper, we propose a neural network-based distributed temperature prediction algorithm including an inter-server joint modeling framework based on the thermal locality principle, which significantly reduces the training time of the temperature prediction model and make the proposed algorithm be easily extended to large data centers. Furthermore, we propose a job scheduling algorithm based on the proposed temperature prediction algorithm. The job scheduling algorithm monitors the server inlet temperature in real time and controls the load of each server using feedback control. It guarantees that thermal reliability of servers and attempts to avoid the creation of a hot point. It selects the best job scheduling strategy based on the result of the temperature prediction algorithm. The two proposed algorithms are evaluated on a small data center. Our results show that the average prediction error of the proposed temperature prediction algorithm is only 0.28 °C in a 10-min predicted field of view. The proposed job scheduling algorithm can achieve approximately 10% cooling energy consumption compared with the load balancing algorithm while ensuring the thermal reliability of the data center.