Research And Implementation Of Operation Data Monitoring And Periodic Anomaly Prediction Method In Cloud Environment

In recent years,cloud computing technology has been developed and mature,and cloud computing technology has been widely used.Cloud computing technology has gradually become the mainstream technology in IT industry.The cloud environment has the characteristics of dynamic,complexity,sharing and large scale.In the complex cloud environment,if the operation data is abnormal,it will affect the normal operation of the cloud environment,and bring great challenges to the availability and stability of the cloud environment.In order to ensure the availability and stability of cloud environment,relevant experts at home and abroad have conducted in-depth research on monitoring,anomaly detection and prediction methods of operating data under cloud environment,and put forward the monitoring framework and anomaly detection and prediction methods of relevant running data,but there are still some problems:1.The data monitoring system in cloud environment has high coupling,low scalability and high resource consumption.2.When the periodic data in cloud environment is detected,due to the lag of periodic anomaly detection,only when the monitoring data is full of a periodic,the data of the whole periodic can be input into the detection model.The results of anomaly detection often fail to meet the requirements of the timeliness of anomaly detection in cloud environment.The current detection results of periodic anomalies can only be feedback on whether the anomaly occurs or not and the degree of anomaly can not be feedback,so it is necessary to predict the periodic anomaly actively and provide more comprehensive prediction results.3.When predicting periodic anomalies,data prediction methods will be used.The existing data predicting methods have some limitations for the accuracy of periodic data prediction.Based on the characteristics of cloud environment and improving the availability and stability of cloud environment,this paper studies the monitoring and prediction methods of running data in cloud environment.The main research contents are as follows:1.In order to overcome some problems in the cloud computing platform running data monitoring framework,and integrate the abnormal prediction function,this paper designs a cloud environment running data monitoring and anomaly prediction framework.2.According to the characteristics of periodic running data in cloud environment,this paper proposed an anomaly prediction method which first predict data and then detects the predicted data to predict the abnormal prediction of periodic running data(A periodic anomaly prediction method combining the attention-based convolutional gating unit code-decoder data prediction model and the encoder-decoder detection model).Firstly,data of the next periodic were predicted by preprocessing the monitoring data and then input the predict data into the detection model to predict whether there was any abnormality and the score of the abnormal degree was given.3.In consideration of the limitations of the existing data prediction methods and the characteristics of periodic running data,the paper proposes using attention based Conv GRU unit to extract the time and space characteristics of periodic running data,and better predict the running data of the next periodic,which lays the foundation for abnormal prediction.4.The prototype system of periodic running data anomaly prediction is implemented on the cloud platform Open Stack.The details of the prototype system are shown to verify the usability of the prototype system.Finally,the relevant experiments are carried out to verify the effectiveness of the method.