Research On Active Forecasting Of Data Center Server Workloads

In the era of cloud computing,the excessive use of data center resources(CPU,memory,disk)and subsequent machine failures have brought huge losses to users and enterprises.Therefore,it makes sense to predict server workload in advance.In the past,research on server workloads mainly focused on trend analysis and time series fitting.We propose a server workload prediction method based on a combined model.First,for data center server CPU utilization time series prediction,a variable-weight Holt-Winter and LSTM combined prediction model is proposed.Use Generalized Recurrent Neural Network(GRNN)to track and predict dynamic weights to obtain weighting factors on different models at different times.Designed and implemented Holt-Winters prediction model,long-term and short-term memory neural network(LSTM)prediction model,differential and moving average autoregressive mobility(ARIMA)model,fixed weight and variable weight prediction model,and clustered with Google data center The data set is used as experimental data.The experimental results show that the average relative error of the prediction of a single model is 11.46%,16.18%,25.13%,11.09% for the fixed-weight combination model,and5.24% for the variable-weight combination model.The variable-weight combination prediction model has stronger applicability.And higher accuracy.Secondly,in order to further improve the CPU load prediction accuracy during the operation of the data center server.The CPU load prediction model integrating the moving average autoregressive model(ARIMA)and the long and short-term memory artificial neural network(LSTM)was established respectively.However,due to the lack of consideration of the original data processing and the limitations of artificial intelligence algorithms in the above models,a combined CPU load prediction method(IEBL)based on isolated forest algorithm(IF),empirical mode decomposition(EMD)and LSTM was constructed.Parameter solving and simulation of the three models built and compared with the experimental results show that the average relative error of the combined forecasting model is reduced by8.71%-18.72%.The experimental results show that the prediction accuracy of the IF-EMD-LSTM combined prediction model is significantly higher than that of any single prediction model,and it has a wide range of application prospects in the field of server CPU load prediction.