Research On Live Migration Of Virtual Machine

Cloud Computing has become a research hotspot of IT field. It has been considered to bethe most important computing and service mode. By using this mode, consumers will be ableto have on-demand access to applications and data in Cloud. In another word, CloudComputing has abilities in providing consumers with various services, such as the requiredstorage, computing resource, platforms and infrastructure etc. The Cloud Computingenvironment providing the infrastructure service needs to construct a huge resource pool andthe project’s core technology is virtualization technology. Thus, virtualization technology isone of the most important foundations of Cloud Computing.With the development of virtualization technology, the research and deploymentfocusing on VM (Virtual Machine) technology have appeared. VM technology has excellentcompatibility, flexibility and security. On the compatibility, VM technology enables multipleoperating systems to coexist on the same physical host. And that each VM is completelyisolated from each other also demonstrates that it has a good security. On the user experience,a VM and a traditional computer are the same. Both can manage own operating systems andapplications independently[1]. However, VMs have a better flexibility which traditionalcomputers have a lack of. VMs can achieve a live migration between physical hosts in thetechnology.In Cloud Computing data centers, the flexibility brought about by live VM migrationtechnology has abilities in addressing some key issues of data centers, such as load balancing,fault tolerance, routine maintenance, energy saving given the biggest concern and so on. Dueto the complexity of computers themselves, the difficulty of live VM migration can beimagined. Live migration of VM is mainly divided into network migration, storage migrationand the running state migration including memory status, CPU status and I/O status. Afteryears of development, network migration technology and storage migration technology aregetting mature. Many companies such as VMware, Microsoft and Citrix etc. have achievednetwork migration in their products. And storage migration has also been achieved byutilizing storage sharing technology such as NAS (Network Attached Storage), NFS (NetworkFile System) and DFS (Distributed File System) etc. However, the running state migration isrelatively backward and still has much room for improvement despite of the rapiddevelopment. What is more, it has huge impact on the whole performance of live VMmigration. How to improve the running state migration, shorten total migration time andimprove the efficiency of migration has been one of the research priorities of current VM migration technologies.This paper has focused on the study of live VM migration mechanism. In this paper, themigrant VM object is divided into two categories: data-intensive and computing-intensive.The related work for live migration of the two types of VMs has been done. Also, live VMmigration policy for the hot issue of energy saving in Cloud data centers has been studied onthe basis of the research of live VM migration mechanism.At the present stage, the research work related to this paper has some shortcomings asfollowing:(1) For live VM migration mechanism, there are three types of mainstream live VMmigration approaches researched widely in today’s international. They are Pre-copy algorithm,Post-copy algorithm and the live VM migration approach based on checkpoint recovery andlog replay technology. These approaches have achieved live migration of VM in multipleaspects, and their variants have also continuously improved the performance of live VMmigration. However, the performance and efficiency of live VM migration still have a lot ofroom for improvement.(2) During migrating a data-intensive VM, Pre-copy algorithm will transmit excessiveredundant memory dirty pages in every round of iteration and thus result in too much datamigrated. Post-copy algorithm easily causes too long total migration time and thus results inthat the occupied computing resource cannot have been released for a long-term.(3) During migrating a computing-intensive VM based on the live migration approach ofcheckpoint recovery and log replay technology, it will reduce migration performance andincrease much data transmitted since the logs generated are excessive and its granularity is toocoarse.(4) In the research of live VM migration policy, for energy saving the current mainstreamplacement selection policy of live VM migration is the random placement selection policy oflive VM migration and the optimal energy-saving placement selection policy of live VMpolicy easily causes a higher failure rate of migration and more energy consumption etc. Theoptimal energy-saving placement selection policy has only optimized the currentenergy-saving effect of Cloud data centers with the result of live VM migration. It is notnecessarily the optimal placement selection policy for energy saving from a long-term pointof view.Aiming to address the above problems, this paper has focused on live VM migrationmechanisms from the two aspects of the data-intensive and computing-intensive; on this basis,this paper has done some related work on the placement selection policy of live VM migrationfor energy saving. The main contribution of this paper is as follows:1. This paper has done summary and survey research work on live VM migrationmechanisms. According to the different technologies employed, the current mainstream liveVM migration approaches have been classified. In these classifications of live migration mechanisms, this paper has introduced and analyzed Pre-copy algorithm and its variants,Post-copy algorithm and its variants as well as the live VM migration algorithm based oncheckpoint recovery and log replay technology in detail. This paper has summarized thecurrent representative live VM migration approaches and analyzed their advantages anddeficiencies. Besides, the future research trends of live migration of VMs have been putforward.2. This paper has done research work on live migration of the data-intensive VMs. Inorder to address the problem that a large number of redundant memory pages need to bere-transmitted since the data-intensive VMs are always updating memory pages frequently,this paper has presented a novel migration algorithm based on hybrid memory copy and deltacompression, which is called HMDC for short. Its main idea includes two parts. One is that itcombines memory pulling copy with memory pushing copy to achieve hybrid memory copy.Hybrid memory copy refers to combining the full memory pushing copy of source VM withthe on-demand copy of target VM to achieve a fast memory synchronization through thismechanism into the process of memory migration, in which source host makes use of memorycache to get XOR delta pages, and then compresses delta pages which are easy to compressby the XOR binary RLE (run-length encoding) algorithm. Source host transmits deltacompression pages instead of dirty pages to target host. By this mechanism, it has increasedmigration throughput and achieved a fast migration. On the XEN virtualization platform,HMDC algorithm has been compared with the classical Pre-copy algorithm and XBRLEalgorithm combining Pre-copy with delta compression. HMDC increases throughput anddecreases total migration data by using delta compression and thus to achieve dirty pagescopy quickly. The three migration mechanisms have conducted several experiments by usingthe benchmarks and the VMs with the representative loads. The experimental resultsdemonstrate that HMDC evidently reduces total migration time, VM downtime and totalmigration data compared with Pre-copy and XBRLE algorithm. It makes the process ofmemory migration more high-effective and transparent.3. This paper has done research work on live migration of the computing-intensive VMs.In order to address the problem that migration performance is degraded and total datatransmitted is excessive since the computing-intensive VMs are always generating much dataof the running state and updating the running state such as CPU and I/O frequently, this paperhas proposed a fast live VM migration approach with CPU scheduling, which is called FMCfor short. FMC is a self-adaptive method which synchronizes the state of the VM migratedbetween the source and target host by log tracing and replaying instead of dirty memory pagestransferring. First, the mechanisms of fine-grained log files division, replay time unitsfeedback and dynamic variable step K are presented. Second, we introduce the CPUscheduling mechanism into FMC to adjust the CPU timeslices in order to reduce the log datasize when the network bandwidth is too low. Experimental results show that the proposed method can significantly reduce migration overheads compared with the classic pre-copyalgorithm in a fast network bandwidth. Identically, the migration downtime of FMC is lessthan that of CR/TR-Motion in a low network bandwidth. The overheads are also acceptableeven in low network bandwidth.4. In this paper, we have researched the placement selection policy of live VM migrationfor energy saving. Since the random migration policy has some problems that the failure rateof migration is high and the total incremental energy consumption is more relatively while theoptimal placement policy has the problem that it may be not enough efficient from along-term point of view, this paper has presented a novel heuristic approach which is calledPS-ES. Its main idea includes two parts. One is that it combines the PSO (particle swarmoptimization) idea with the SA (simulated annealing) idea to achieve an improved PSO-basedTheory and Mathematical Statistics and once again utilizes the SA idea to deal with the dataobtained from the improved PSO-based process to get the final solution. And thus the wholeapproach achieves a long-term optimization for energy saving. PS-ES achieves a longer-termenergy saving as it has considered not only the optimization of the current problem scenariobut also that of the future problem. The experimental results demonstrate that PS-ES evidentlyreduces the total incremental energy consumption and better protects the performance of VMrunning and migrating compared with randomly migrating and optimally migrating. It makesthe result of live VM migration more high-effective and meaningful.