Container Cluster Management Method For Large-scale Heterogeneous Evaluation Tasks

In recent years,the virtualized container cluster represented by Docker has gradually become the core technology for all kinds of software enterprises to support the upgrade of large-scale business systems.In the new intelligent IT teaching and training platform,in order to support the training and automated evaluation of large-scale heterogeneous IT technology,building an evaluation cluster based on virtualized containers can effectively isolate the differences between the IT programming language and the technical architecture and provide teaching and training cloud services for the full IT technology stack.At the same time,it can effectively improve the hardware and software resource utilization and user experience of the training service.Virtualized container clusters are the foundation for efficient automated evaluation services.For the highly concurrent heterogeneous IT technology evaluation task set,providing isolated,customized,and highly available evaluation services involves two aspects: First,for different IT technical architectures and programming languages,it is necessary to provide different docker images.There are so many kinds of IT technologies and programming languages,so the types and numbers of Docker images that need to be built are very large.How to effectively manage large-scale images becomes the first key problem to be solved.Secondly,when online evaluation is conducted by large-scale users.It is necessary to ensure that the background hardware resource utilization is improved while responding to user requests in real time.To this end,dynamic scheduling and management of the container's lifecycle based on user application scenarios is the second key issue that needs to be addressed.Aiming at large-scale heterogeneous image management,the paper focuses on the root causes of low image multiplexing rate and image out of control growth,and explores solutions through online data analysis of IT teaching and training platform,then conducts an experimental demonstration.The paper proposes a method based on mirror depth retrieval strategy to improve the success rate of image location,as well as image de-redundancy and image merging algorithm based on image similarity analysis,and optimizes the organization of images through service-image plug-in and separation of registries.For the scheduling management issues of container clusters,the paper focuses on the scheduling algorithm and resource allocation strategy of container.For each life cycle stage of the container,the paper analyzes the main contradictions of container cluster application at different stages,and proposes relevant scheduling algorithms based on problem modeling to solve the problem of container cluster scheduling and control under the scenario of large-scale heterogeneous evaluation task.Related algorithms include: improved cluster node selection algorithm,flow control algorithm,adaptive container runtime state transition strategy,and resource floating quota method.The task scenario of evaluating is actually a typical cloud platform task scenario.These container scheduling and management methods have certain reference significance in the cloud platform construction.