A Out-of-Order Parallel Framework Research And Implementation Based On Docker Container Dynamic Migration

With the rapid development of cloud computing technology,the virtualization technology has the characteristics of isolation,simplicity and hardware independence,which makes it gradually become the foundation of the cloud platform to provide services.The cloud computing technology integrates software and hardware resources,and provides storage and computing functions such as object storage,capacity,and performance in a virtual environment and a bare metal environment.As the user demand increases,the virtualized environment gradually replaces the bare metal.surroundings.At present,virtualization technology is mainly divided into two categories: virtual machines and containers.Cloud service providers use these two technologies to virtualize resources(CPU,memory,network,read and write functions,etc.)in the cloud.In order to provide users with stable services and reduce service energy consumption,cloud service providers not only need to ensure low overhead and high scalability of resource management in cloud computing environments,but also provide load balancing and error tolerance of servers.Live Migration is not only one of the important guarantee technologies for efficient and stable cloud platform services,but also helps to solve problems such as load balancing and error tolerance between servers.The containerized virtual machine technology belongs to a completely isolated process group inside the operating system.The biggest difference from the traditional virtual machine is that it has extremely low management overhead.The traditional virtual machine provides all the independent resources and manages the resources through the hypervisor,which brings high overhead.The container realizes the management and allocation of resources by directly calling the operating system kernel.The resource management overhead is far lower than that of the traditional virtual machine.From the aspects of energy consumption,isolation and practicality,the container gradually becomes more and more users and The choice of cloud service providers.Docker container is one of the most mainstream technologies in many container technologies.It is an open source and lightweight container engine project.The biggest feature is low management overhead and easy deployment.However,Docker officials currently do not support the dynamic migration of containers,and the industry has less research on the dynamic migration of containers.This paper comprehensively considers the characteristics of resources in the cloud and Docker itself,and uses the pre-existing technology,Pre-copy algorithm and container C/R operation to design and implement the cloud platform-based out-of-order parallel Docker container dynamic migration architecture,making Docker container technology The cloud platform has better service functions.The experimental data shows that the dynamic migration architecture proposed in this paper has better migration convergence,less data transmission and service degradation rate than the traditional container and virtual machine dynamic migration in different scenarios.The main contributions of this paper are as follows:1)This paper designs a dynamic migration algorithm based on out-of-order parallel Docker container,which adopts cloud pre-storage manager,Pre-copy algorithm and checkpoint/restore(C/R)technology.In this way,the network and hardware resources of the cloud platform are fully utilized,the downtime is reduced,the transmission volume of the source host in the dynamic migration phase is reduced,and the low overhead and practicability of the Docker dynamic migration are realized.2)Implemented a Docker container dynamic migration architecture based on pre-existing technology and container C/R operations.In order to further reduce the data components in the source host and make full use of the characteristics of the cloud platform resources,this paper adopts the dynamic pre-existing technology of dynamic control and pre-stored anti-products based on the dynamic migration of Docker containers.For different types of applications that users run on the cloud platform,the threshold value of the pre-stored phase threshold can be set according to requirements.The traditional container technology can use the bare metal C/R operation to obtain the memory page of the internal application of the container to achieve dynamic migration.However,due to the special function of the Docker daemon,the bare metal C/R operation will cause the Docker daemon to not sense the change of the container.Therefore,it cannot be successfully recovered during the shutdown phase.Finally,this article uses the libcontainer library of the Docker container to directly call the CRIU function in the Linux kernel to ensure that the daemon participates in the entire migration process.3)Design and implement an optimization scheme based on data transmission.The performance metrics of the dynamic migration are analyzed separately,and the balance between Downtime,TMT and TDT is combined.Finally,the combination of real-time dirty page thresholds and set-bound loops is used to optimize the performance of Docker dynamic migration.