Workload Prediction and VM Clustering Based Server Energy Optimization in Enterprise Cloud Data Center

The abstract should briefly summarize the contents of the Server energy consumption of data center is an important issue of energy management. Energy optimization of server is also necessary to reduce energy consumption of data center cooling and power supply, and reduce the operation cost of whole data center. High server energy consumption is mainly caused by excessive allocation of IT resources according to the highest application workload. This paper studies the optimization algorithm of server energy consumption in enterprise cloud environment. By introducing deep learning model LSTM to predict application workload, the proposed algorithm can dynamically determine the starting up and shutting down time of virtual machines (VMs) and physical machines (PMs) according to the workload to realize the matching of application workload needs between IT resources. K-mean clustering algorithm is used to find VMs with similar starting up and shutting down time and put them on same PM group. By properly extending the running time and increasing number of VMs, the algorithm can compensate the impact of inaccurate prediction and workload fluctuation and guarantee the applications QoS. The simulation results show that the proposed method in this paper can reduce the energy consumption of servers by 45-53% with QoS guarantee when the prediction relative error is 20%, which can provide a good balance between energy saving and application QoS.