A Research On Live Memory Migration Of Virtual Machines Based On Pre-copy In Cloud Computing

Computer hardware resources can be abstracted by virtualization technology.The abstract resources are redistributed to improve the resource utilization of the computer physical hardware.Live migration technology is a great important technology in virtualization,which migrates virtual machines from source hosts to new destination hosts almost without downtime,avoids the impact of the long downtime on applications running on virtual machines.However,the business requirements of the migration process are changed at any time in the cloud platform,resulting in the generated data are written into memories,increasing the difficulty of memory migration,affecting the performance of live migration and applications,and ultimately affecting the cloud platform to provide services.Therefore,researching on the live migration of virtual machine is necessary for cloud platforms.Pre-copy-based memory copy method is widely used in various mainstream virtualization platforms.For example,the memory copy in KVM use pre-copy method as default.Under low load condition,the live migration of virtual machine can proceed smoothly.When the load in a high level,the dirty page rate will be high.Due to the insufficient physical bandwidth,the modified memory pages cannot be transmitted to the destination host in time and gradually accumulate,resulting in the failure in live migration of virtual machine.To solve the failure in live migration of virtual machine under high load condition,this thesis improves on the basis of analyzing the stop-and-copy condition and the causes of dirty pages generation,mainly aiming at reducing the generation and the transmission of dirty pages.When the dirty page's convergence rate begins to decrease,the running time of CPU will be reduced to decrease the amount of data written to memory.By limiting the activity of CPU to reduce the production of dirty pages per unit time,the generation rate of dirty pages will less than the transmission rate of bandwidth,the migration can proceed smoothly and the iteration time will be decreased.By recording the historical data of memory dirty pages to judge whether the memory pages are frequently modified,this thesis modifies the copy strategy of memory pages,the frequently modified memory pages will not be transmitted in this round.Finally,the total migration time and the transmission data will be decreased.This thesis builds an experimental platform based on KVM to verify the feasibility of limiting the activity of CPU and reducing the retransmission of dirty pages.The final experimental results show that the improved algorithms proposed in this thesis perform well in reducing the migration time and the amount of transmission data.The CPU's running time is reduced when the convergence speed of dirty pages is dropped.Therefor,dirty pages can also converge at high speed under the high load condition.Finally,the downtime is reduced by decreasing the dirty pages in the last round.