Docker Container Anomaly Monitoring Based On Triple Exponential Smoothing And LSTM Mixed Model

With many advantages,Docker has become one of the mainstream technologies in cloud computing.Therefore,the stable and reliable operation of Docker containers has become a key point in the field of cloud computing,and the current container monitoring technology that mainly uses manual threshold setting is difficult to put into use in the case of a large number of containers and a variety of application types in the container.In addition,the real-time nature of the container abnormality monitoring system also has high requirements,and the reliability of the container must be able to locate the abnormality in the container in a timely manner.According to the difficulties of the current Docker container anomaly monitoring technology,anomaly monitoring is performed on the Docker container resource metrics based on the predicted anomaly detection algorithm,and Docker container anomaly monitoring system based on Triple Exponential Smoothing and LSTM(Long Short Term Memory)mixed model is implemented.Considering the real-time nature of Docker container anomaly monitoring,the three-exponential smoothing model and the LSTM model are designed as online learning prediction models,while taking into account the impact of abnormal points on the prediction model,the updates of these two models are optimized for abnormal points Finally,the two models are combined with improved MAPE(Mean Absolute Percentage Error)method to achieve efficient and accurate prediction of container resource metrics.A dynamic sliding window that adjusts the size according to the monitoring status is used to store the prediction error,and the prediction error is modeled as a normal distribution,and whether the monitoring data is abnormal according to the 3? principle is determined.Experiments show that the system can efficiently predict and anomaly detect Docker container resource measurement data.The prediction accuracy of the Docker container resource measurement is improved by an average of 13.77%.For the Docker container anomaly detection,the F-Score is increased by an average of 31.16%,and the anomaly detection overhead In about 5.6ms,it meets the real-time requirements of Docker container abnormal monitoring.