Optimization And Implementation Of Virtual Machine Dynamic Migration Algorithm

The virtual machine dynamic migration technology enables the virtual machine to move on the compute nodes of the cloud data center,and the services running on the virtual machine are basically not interrupted during the whole process,so that the maintenance cost of the cloud data center is greatly reduced.However,when a large number of memoryintensive applications are running in a virtual machine,memory pages are frequently read and written.The existing dynamic migration algorithm will repeatedly transmit a large number of dirty pages,which not only increases the time overhead of dynamic migration,but also makes migration.The time is extended,and it also takes up a lot of computing resources,affecting the normal operation of the virtual machine itself and other virtual machines of the same host.In order to improve the migration efficiency of virtual machines,Firstly,the paper analyzes the shortcomings of KVM(Kernel Virtual Machine)'s native pre-copy algorithm,and proposes an algorithm based on locality principle to classify virtual machine memory pages.Based on this,an improved memory pre-copy algorithm(LPBP)is proposed to reduce the number of re-transmissions of high dirty pages to reduce hot migration time and downtime.Secondly,in order to solve the problem that the pre-copy algorithm cannot converge under the condition of excessive load,an improved virtual machine shutdown judgment algorithm is proposed,which not only judges the shutdown condition from multiple performance indicators,but also automatically cools the vCPU when the load is too high.Thirdly,the LZ4 compression algorithm is introduced to further optimize the memory migration process,reducing the total amount of data transmitted during the entire migration process.Finally,this paper implements the above optimization scheme on the KVM platform and tests its performance,which is compared with the native pre-copy algorithm and a typical memory migration pre-copy algorithm TPO.The experimental results and analysis show that the optimized memory heat migration strategy is effective,which greatly shortens the total migration time and downtime,reduces the total transmission volume of dirty pages,and performs better under memory-intensive load.In a high memory load environment,the LPBP algorithm reduces the downtime by about 42% compared to the native algorithm,about 54% shorter than the TPO algorithm,and 10% faster than the Xen-alike algorithm.In addition,the total migration time and the amount of data transferred are greatly reduced compared to the other two algorithm pages.