An adaption scheduling based on dynamic weighted random forests for load demand forecasting

Abstract

With the development of cloud computing, energy consumption has become a major and costly problem in data centers. To improve the energy efficiency of data centers, we analyze the influence factors of energy consumption and discover that reducing the idle servers can effectively cut down the energy consumption of data centers. Then the load demand forecasting algorithm using weighted random forests is proposed. And time factor matching coefficient obtained by considering the day type and the time span is employed to calculate the weights. To enhance the forecasting performance, an error correction strategy is also introduced into the forecasting model. The experimental results show that these strategies further improve the prediction accuracy, and the root-mean-square error is 2.6–4.1% lower than other forecasting algorithms. We finally design an adaptive scheduling technology that utilizes short-term prediction of load demand. This technology adaptively adjusts the scale of the data center cluster based on the forecast results. The simulation results indicate that the technology can reduce 12.5% energy consumption while ensuring the service quality.