Design And Implementation Of Application Container Performance Monitoring And Scheduling Migration Scheme Based On Docker

Due to the advantages of lightweight virtualization of Docker containers,more and more companies use Docker Swarm as their main task execution environment.Usually,many different types of application containers are deployed on each node in Docker Swarm to perform different tasks.When a large number of tasks are running and scheduled at the same time,it is inevitable that abnormal conditions such as resource preemption and excessive load will occur,resulting in unbalanced cluster load.In this case,it is very necessary to monitor the running status of the cluster.For Docker Swarm,the native cluster of Docker containers,although it has a default container scheduling strategy,it does not have a complete set of anomaly discovery,analysis and resolution programs.In response to the above problems,this thesis conducted the following research.Firstly,based on the requirements of container cluster node monitoring and container performance monitoring,this thesis applied Prometheus,c Advisor and Grafana technologies to monitor Docker Swarm in real time.Secondly,in view of the different resource usage preferences of application containers,this thesis proposed a classification model that combines PCA dimensionality reduction and KNN algorithm,which divides application containers into four types: CPU-intensive,memory-intensive,disk I/O-intensive,and network-intensive.Container performance data is collected by monitoring services and used as a data source for multiple experiments.The experimental results show that compared with the existing KNN classification algorithm,the classification accuracy of the algorithm combining PCA dimensionality reduction and KNN is improved,and the running time is greatly shortened,verifying that the feasibility of applying this classification model to the container scheduling migration scheme.Thirdly,by studying the existing virtual machine scheduling migration strategies,aiming at the four stages involved in the virtual machine scheduling migration strategies: migration timing,selection of the virtual machine to be migrated,target node selection and migration method.Combined with the features of Docker container,this thesis proposes the corresponding algorithm strategies for container migration.At the same time,the classification model is applied to the selection strategy of containers to be migrated,and a minimum resource utilization container selection strategy based on the combination of PCA dimensionality reduction and KNN classification is proposed called PKMU(PCA-KNN-based Minimum Utilization Container Selection Policy).A container scheduling migration scheme is designed and implemented.Finally,the monitoring solution and container migration scheduling solution are integrated into a set of automated container cluster monitoring and migration solutions,and tested in the Docker Swarm environment.The experimental results show that the solution can monitor the container cluster in real time,and under the condition of unbalanced load of the cluster,the load balance of the cluster and the resource balance of the nodes are improved through the automatic migration of the containers among the nodes,which fully proves the research significance and practical value of this topic.