Cloud Server Aging Analysis And Prediction Based On Load Change

Software defects that are undiscovered or difficult to repair inevitably lead to anomalies and errors in the software system.With the long-term accumulation of exceptions and errors,the performance of the software system gradually degrades,and the frequency of exceptions and errors gradually increases.The phenomenon of unexpected crash is called software aging.Cloud servers are widely deployed in various industries to provide users with stable and reliable services for a long time.However,cloud servers are prone to accumulate a large number of anomalies or errors under long-term workloads,resulting in cloud server aging and affecting system reliability and availability.It is difficult to directly extract the aging features of server performance and resource status under load changes.To analyze the software aging of the cloud server under load changes,analyze the aging trend and aging status of the cloud server in terms of single aging indicator and multidimensional aging indicator.In terms of a single cloud server aging trend of aging index analysis,first of all,through the iterative aggregation algorithm divided cloud server running state,and then through the analysis of the trend test method performance indicators of aging index and resource-based aging trend,cloud server aging trend to analysis,determine the cloud server aging phenomenon.In terms of multidimensional aging and index analysis of cloud server aging state,firstly,the cloud server aging state is divided based on K-means clustering algorithm,and then the cloud server aging source is realized based on Convolution Neural Network(CNN).Select an aging indicator that best reflects the aging status of the cloud server,that is,a key aging indicator for subsequent aging prediction.Based on the traceability results of aging analysis,key aging indexes were selected for aging prediction.In view of the nonlinear,tendency,periodicity and randomness of the operating environment of cloud server system,the aging index Trend characteristics and cycle characteristics are extracted based on the periodic Trend decomposition method(STL).In this paper,the regressive Integrated Moving Average model(ARIMA)and Genetic Algorithm(GA)were applied to calculate the regressive Integrated Moving Average model.GA)optimized Long short-term Memory(LSTM)combination software aging prediction model STL-GALSTMARIMA.The combined model mainly consists of three parts: first,the characteristics of the aging index are extracted based on the period and trend decomposition method,and the trend characteristics,period characteristics and residual are obtained.Then,ARIMA model suitable for stationary linear data was used to fit the trend characteristics of the aging index,and LSTM model suitable for non-stationary nonlinear was used to fit the period characteristics and residual of the aging index.Finally,the final prediction results are obtained by STL inverse transformation of predicted values of each model.The experimental results show that the aging prediction model STL-GALSTM-ARIMA is superior to ARIMA,LSTM,SVR and random Sen aging prediction models.Root Mean Squared Error(RMSE)and Mean Absolute Error(MAE)are improved in different degrees.Rejuvenation interval is predicted based on preset aging threshold.Based on the premise that software aging in cloud servers will lead to the gradual decline of system performance and reliability,referring to the concept of Remaining Useful Life(RUL),An aging prediction method based on Support Vectors(SVs)and Gaussian function fitting(GFF)is proposed.A gaussian function fitting model based on density clustering was established based on the statistical feature data of the aging index sparse by support vector regression.Based on the optimal aging curve,the RUL of the system before reaching the aging threshold was evaluated to predict when the aging of the system occurred.Compared with other RUL prediction methods,the cloud server rejuvenation interval prediction method also has advantages in precision and convergence speed.This paper also designs and implements the cloud server aging analysis and prediction platform,and realizes the aging analysis and prediction of cloud server combined with the aging analysis method and aging prediction method proposed in this paper.The system integrates modules such as ECS monitoring,aging analysis and aging prediction,and can analyze and predict the aging of ECS under different loads.