Container Image Placement And Caching Strategy Based On Ceph Distributed Storage

With the development of cloud computing technology,the types of deployment scenarios and container clouds have become more diverse.However,the volume of container images is also increasing and the deployment efficiency of the container has been negatively affected.In the container cloud platform,container deployment efficiency and service quality have greatly exert effects on user experience.Therefore,how to deploy containers efficiently and low latency has become a hot topic in current research.It has been proposed that container deployment is divided into two stages: image pull process and image start process.Research shows that about 76% to 92% of container deployment time is spent on image pull process.As a typical representative of container cloud engine,Docker has been widely used and deployed.For this reason,our study deeply explores the deployment strategy of Docker native containers and optimized the efficiency of image pull for Ceph distributed storage.The specific research work is as follows.Firstly,this study aims to explore the placement strategy based on the attribute weight of Ceph atomic image.Docker's native strategy is unable to allocate resources reasonably according to the characteristics of the image itself,frequency of use,and other attributes,resulting in low resource utilization and long delay in pulling the image.Therefore,this research proposes a placement strategy based on the attribute weight of Ceph atomic mirroring.This strategy optimizes the node placement strategy by establishing the atomic image attribute weight model and the Ceph resource model,ensuring that the atomic image with frequent use and high reuse rate has sufficient resources.Besides,this study made an experimental comparison between the native strategy and the paper optimization strategy.The results show that the placement strategy based on the attribute weight of the Ceph atomic image can improve the resource utilization of the server,reduce the image pull delay,and speed up the efficiency of container deployment under the different conditions of sufficient and scarce server resources.Secondly,by analyzing the Ceph file read and write process,it was found that insufficient disk I/O bandwidth on the server side can easily become a bottleneck when image is pulled.To clear up this problem,a caching strategy based on the relevance of atomic images is proposed.This strategy analyzes the relevance of the downloaded atomic image,and combines the cost of the atomic image in the cache to replace the cache on the server.In addition,the caching strategy and the atomic mirror placement strategy are combined to conduct comparative experiments when the server-side disk bandwidth is under different loads.The results show that the use of thesis caching strategy can effectively reduce the client's disk I/O bandwidth requirements on the server and reduce the delay of mirroring.This paper has indentified that both the placement strategy based on the attribute weight of Ceph atomic images and caching strategy based on the correlation of atomic images can improve server resource utilization and effectively reduce the deployment delay of container images.