Research And Implementation Of A Docker-based Private Cloud Host System

With the development of cloud computing,cloud host is used more and more widely.Most of the cloud computing providers have provided cloud host rental services under the public cloud.The service has also been favored by the users from all walks of life.The users can lease services on demand,and adjust capacity of resources flexibly according to business needs.Those services can help users to build projects quickly,deploy applications securely,reduce the difficulty of development and operation and overall IT costs.Most cloud hosts provide a standardized service for the public,and can't maximize its advantages in some specific circumstances,in addition,most cloud hosts consider Xen or KVM as their virtualization technology,which are thought to be "heavy".The emergence of Docker container technology,blurs the strict boundaries between the PaaS and IaaS gradually.The characteristics of its lightweight virtualization gradually translate the traditional application development and operational mechanism into a new way.This thesis mainly studies and implements a Docker-based private cloud host system,which is mainly provide services for the research and development team of project.This thesis analyzes the requirements of the cloud host system,classifies the system according to the requirements,and designs the architecture of the Docker-based cloud host system in detail.On this basis,this thesis analyzes the built-in scheduling strategy of Docker Swarm and summarizes their shortcomings.In response to these shortcomings,the corresponding solutions are given.In this thesis,these solutions are integrated and a multi-dimensional dynamic weight scheduling strategy is proposed,which not only considers the all kinds of resource usage of nodes,but also considers the resource that the container to be dispatched emphasis on.The proposed solution can balance the cluster nodes in the use of resources and improve the stability of the system.Secondly,according to the layered storage method of Docker image and container,a migration scheme based on private warehouse is proposed to shorten the total time of migration,reduce the amount of transmission data and improve the efficiency of migration.Then,in order to fit the Docker container's micro-service thinking,each container runs only one service process as much as possible and presents a multi-level authentication solution that are based on the fortress architecture,which not only meets the requirements of multi-client login,but also avoid to install additional ssh service in the container.At the same time,the solution reduces the complexity of remote login management.Finally,using the advantages of Docker containers and mirrors,an easy-to-extend,continuous delivery model is proposed.In this mode,according to the actual needs of the project,the system can horizontally extend new container that are used to build the code package of the project,Handle dependencies of the project and automate the deployment.This solution can increase the flexibility of continuous delivery,and help the project team quickly deploy projects to improve productivity.In addition,this thesis tests the four modules of load scheduling,migration of containers,remote login and continuous delivery,including functions and performance.It proves that the solution proposed in this thesis not only reduces the cumbersome workload of project and improves work efficiency,but also balance the use of resources of the nodes in cluster and improve the efficiency and stability of the system.