Research Of I/O Consumption Performance Optimization For Containerized Relational Database Service

The integration of traditional relational database has some challenges to store large-scale data,as traditional relational database are limited in computing and storage.With the rapid development of Docker container technology,the containerized environment combined with Docker and Kubernetes has the advantages of automated management and elastic scaling,which benefits for scaling data and changeable managements.Therefore,containerized deployment of Relational Database Service is an inevitable trend.However,containerized RDS(Relational Database Service)has I/O performance consumption(response delay),which causes the RDS capability to decrease QPS(Queries Per Second).Therefore,this thesis proposes an optimization method based on containerized shared cache model and an optimization method based on automatic scaling mechanism.The main research contents of this thesis are as follows:(1)A design scheme of a shared cache model based on a containerized environment is proposed.The I/O optimization solution for the traditional computing and storage separation architecture is expensive and the optimization solution is only for the computing layer or storage layer.Refer to the traditional RDS application scenario.In order to optimize I/O,the distributed is used to intercept a large number of read and write requesteds to avoid direct I/O with the storage layer.Design a shared cache model based on a containerized environment to optimize the I/O consumption of RDS persistence process.The design process comprehensively considers the characteristics of the resource objects Pod,Namespace,and Service of the containerized environment,and designs the availability,consistency,uniqueness of the Key value,and the serial access mode of the shared cache model in detail.The shared cache model designed in this thesis reduces the I/O response latency of containerized RDS in a separate computing and storage architecture.Also it,improves the performance(QPS)of containerized RDS and provides acceptable service capabilities.(2)An improved automatic scaling algorithm IMHPA(Improve Horizontal Pod Autoscaler)is proposed.Kubernetes' own threshold-based HPA(Horizontal Pod Autoscaler)algorithm has problems of unresponsiveness and the existing prediction-based HPA algorithm has high probabilities of prediction errors.The IMHPA algorithm proposed in this thesis supports a variety of indicator types(Utilization,Average Value,Value)for timely and accurate expansion / reduction,thereby increasing the throughput of the shared cache model and reducing I/O response time for containerized RDS.(3)Designed and implemented a containerized RDS performance optimization prototype system.The prototype system includes cluster information submission,RDS registration,shared cache model,auto-scaling model,Pod workload and RDS performance demonstration.Provide an efficient relational database random registration and RDS perform real-time display platform.In summary,this thesis reduces the I/O consumption of the containerized relational database by designing a shared cache model based on containerization and an automatic scaling mechanism based on the IMHPA algorithm.The proposed method effectively reduces the I/O response time of RDS.Also,it designed and implemented a containerized RDS performance optimization prototype system including RDS registration,shared cache model creation and automatic scaling strategy.