Studies And Implementation Of Live Migration On Docker

Compared with the traditional virtualization technology,lightweight virtualization technology can offer a more convenient and easier mode for virtual machine management and deployment.Containers,as one typical representative of the lightweight virtualization technology,can promote the development of the PaaS level deployment,contributing to the expansion of cloud deployment as well.Under this background,this paper studied the live migration algorithms for traditional virtual machines and the Docker container technology in depth.For live migration algorithms of traditional virtual machines,this paper studied the Pre-copy algorithm,the Post-copy algorithm,several improved algorithms based on Pre-copy,and so forth.For the Docker container technology,this paper studied the system architecture and the working principle of Docker container,and Docker's three key techniques especially,namely AUFS,namespaces and cgroups.Based on this,the paper proposed a live migration algorithm for Docker containers through iterative memory duplication from the source host to the destination host,which can both improve the migration efficiency and reduce the downtime greatly.After this,the paper presented several simulating tests for the Docker live migration algorithm.Five different scenarios were adopted during the experiment,including 1)idle state,2)static Web server,3)dynamic Web server,4)Unixbench and 5)memory-intensive workload.The experimental results show that:In terms of the migration time and downtime,compared with the traditional virtual machines,the performances of the first four cases were greatly improved,while the performance of the last case was not improved obviously,but there was still some progress.In terms of the bandwidth resource usage,all the cases occupied a relatively low share for most of the time during the migration process.Only in the last few rounds of the memory duplication,a relatively high share of the bandwidth resource was occupied,but this period of time lasted very short.On the whole,the amount of the network resource used was within a reasonable range during the whole migration process.In terms of the impact on the quality of service and performance,the first four cases were slightly influenced,but the last case was influenced greatly.