A Strategy Of Live Migration Based On Xen Virtual Machine

Virtualization uses a kind of isolation technology to reduce the underlyingheterogeneity of cluster in the cloud computing environment, improves resourceutilization according to virtual resources designed in the cluster. It is one of the coretechnologies of cloud computing, and is the current research focus in the cloudenvironment. The live migration of virtualization enables seamless migration ofvirtual machines, migrates the virtual machine from the source physical host to thedestination physical host when running the business, and the safety and efficiency ofmigration. At present, the dynamic migration performance is improved by thespecific applications of the SaaS layer, and this way improves the adaptability of thecross-platform, but significantly reduces the dynamic migration performance.For the Xen platform, this paper combines the practical application scenariosand Xen underlying migration and proposes a strategy of live migration based onXen virtual machine. Specific works are as follows:Firstly, in order to migrate the virtual machine in the cloud computingenvironment, Xen compares the send-page bitmap with the skip-page bitmap whenmigrating and determines whether memory pages are retransmitted or not. Aiming atthe problem that multiple retransmissions increase network transmission overheadfrom the comparison, this paper proposes a memory optimization algorithm based onAR model. According to all recorded intervals when a memory page are modified,the new algorithm predicts the next modification time of the memory page andretransmits the memory page when the next modification time is greater than athreshold. The optimization algorithm shortens the migration and network overheadof the virtual machine.Secondly, on the basis of the AR model, we analyze of the situation leading tolower quality of services when shortening the time of migration. ForRATE_IS_SAVE mode, it is maximizing the allocation of network bandwidth thatbrings the insufficient bandwidth reserved. We optimize the original model using the self-adaptive algorithm, and propose the network bandwidth allocation strategybased on exponential smoothing algorithm, and have a reasonable forecast of thereserved bandwidth on the rate of the dirty pages in the iterative process, thenallocate the remaining bandwidth to the migration. This way shortens the migrationtime effectively and ensures the quality of service migration.Finally, we collect the experimental data from memory optimization algorithmbased on AR model, network bandwidth allocation strategy based on exponentialsmoothing algorithm and combination of the first two studys, and analyze theperformance indicators about migration time, downtime and service quality. Theresults show that the proposed algorithms have shortened the migration time andensure the quality of service migration.