Research On Deployment Of Virtual Machines On Cloud Platform

Cloud computing is at the forefront of information technology. Cloud platform provideslots of computing resources and software resources. In order to provide better infrastructureservice and to improve the utilization rate of resources, virtualization technology has beenapplied in the cloud-computing field, through which, cloud provider can offer servicesaccording to users’ needs. Because of the development of the technology like Vmware, Xen,some researchers and industrial enterprises have got innovation progress on cloud computingand its applications. In the wake of development in virtualization technology, underlyinghardware furnishes more and more support for virtualization. Virtualization technologybecome the important basis of cloud computing.Virtual machine (VM) technology, as the most important part of virtualizationtechnology, has already been used widely in cloud computing. User experience of operatingon physical machine is the same as operating on a virtual one, so through virtualizationtechnology, we can devise one physical machine into several individual “ones” on the cloudplatform. People can set up several independent VMs on one physical machine, then installdifferent operating systems, services and applications. This makes virtualization technology tobe more compatibility, flexibility and security in cloud computing application.These strengths of VM can also be brought out in its deployment, migration andmanagement under cloud computing environment. Some key issues of cloud computing, likeload balancing, task scheduling and fault-tolerant etc. are partially resolved because of VMtechnology. The “green cloud” aimed at energy saving of data center, which attractstremendous attention, also need the technology related to VM. Lots of VM techniques havebeen applied intensively on cloud computing, but comparing to other issues, more researchesand optimizations on the deployment, migration and management of VM have yet to be done.These issues seriously impact the whole performance of cloud platform. How to customizeVM image efficiently, to reduce deployment, migration time and to deploy and migrate VMreasonably are the main issues in current research of cloud computing.This paper is focused on the customization, deployment strategy and deploymentmechanism of VM in cloud computing. Some researches related to the subject are alreadystarted, but they still need to be optimized and worked on.(1) In the matter of customization of VM image, considering of the security, most of theVM images used Linux kernel. However, the fact that traditional VM images are too big andtake a lot of system resources. It will impact the performance of the deployment process, sothe images need slim down. Linux kernel tailoring is the main solution to reduce the size of operating system in today’s world. However, Linux kernel tailoring takes more time tocustomize under cloud computing environment. It mainly used in embedded system and someparticular domain, besides, users’ requirements are different from each other under cloudcomputing environment. Therefore, it can’t satisfy users’needs flexibly and it isn’t convenientto be applied in customization of VM image under cloud computing environment.(2) On the issue of the placement strategy of VMs deployment, most of currentresearches didn’t pay attention to the placement problem in the process of VMs deployment.They used relatively easy strategy in deployment, and only focused on the location of VMmigration after deployment. If they had done, the migration frequency of cloud center wouldbe greatly reduced, so that the energy consumption could be reduced too. There are manyplacement strategies to select the VMs place, among which heuristic algorithm is widely usedin this kind of issue. But when cloud platform have a certain size, and a number of VMs todeploy, traditional heuristic algorithm is difficult to find a good place efficiently with anaffordable system expenses.(3) In the deployment mechanism, dynamic migration mechanism is the main method.There are three mainstream mechanism of VM migration: Pre-copy algorithm, Post-copyalgorithm and the hybrid memory copy algorithm which mixes the previous two. All thesethree methods can realize the deployment of VMs and many improved methods have beenproposed to increase the performance of deployment. But there is still a lot of improvementspace in deployment performance and efficiency. Pre-copy algorithm must transfer mountainsof redundant dirty pages in iterative process, which will cause the process to transfer moredata and to take more bandwidth. Post-copy algorithm maybe cause long downtime whichwill delay the whole deployment time and take more computing resources. Mixed memorycopy algorithm with Pre-copy and Post-copy can make up the shortfall of the first twomethods in some degree, but deployment performances still need improvement.Aiming at the issues presented above, this paper studies the VM image customizationwith LFS. We study and design a placement strategy of VMs deployment which is sensitive toenergy consumption. Besides, it improves the hybrid memory copy algorithm in combiningwith the incremental compressing method and uses multicast technology to improve theefficiency of deployment process. Main contributions are as follow:1. Using LFS technology to customize VM image according users’ needs. We create userinterface through which users can send their requirements information. Cloud platformprovides management interface of applications and system components, moreover, it generatethe script program related to the automatic installation of applications and system components.Cloud platform generate the configuration file after receiving users’ information. Thecustomization process of VM image search the minimum matching image according to theconfiguration file and do the incremental installation. The obtained image is the exactly onerequired by users and it will be saved on cloud platform for future customization uses. Theuse of LFS greatly reduces the system size and the resources consumption on the premise that users’ requirements are satisfied. Moreover, the boot speed has been seriously raised while thesystem cost and network overhead significantly decrease.2. Taking performance per watt as the standard for evaluating the placement strategy ofVMs deployment. The placement strategy in traditional VMs deployment process didn’tconsider the performance and energy consumption issues of cloud center. This paper proposesa new distributed parallel genetic algorithm (DPGA) for deploying a number of VMs inlarge-scale cloud center by using DVFS technology, meanwhile taking QoS of users andenergy saving of cloud platform in consideration. This algorithm can be separated into twostages. Firstly, considering the coverage rate of solution space, we select initial populationsdispersedly and averagely. Then the genetic algorithm executes parallelly and distributedly onthe selected hosts. The algorithm executing in each host will get an optimal solution. Secondly,we take the solutions obtained in the first stage as initial population and begin the geneticalgorithm procedure of second stage. The optimal solution obtained in the second stage is thefinal solution of DPGA. Through DPGA, cloud center can reduce energy consumptionsignificantly in condition of users’ QoS. In addition, it can ensure the high efficiency underdifferent demand.3. Research on dynamic deployment mechanism of VMs. We proposed a new VMsdeployment mechanism DCVM with the combination of pre-copy algorithm and post-copyalgorithm. First of all, copy the metadata from the source VM to the target VMs. Then thesource VM pushes memory pages to the target VMs at regular time, the target VMs takesmissing pages from source VM during its boot process. Finally, the source VM stops to pushall memory pages remaining. DCVM transfers metadata and memory pages by multicasttechnology. Multicast technology can transfer data to all target VMs at once and can realizepage prefetching of target VMs. The incremental compressing technology is used in thememory pages transfer process for incremental compressing the dirty pages. Firstly, deal withthe dirty pages with the temporary saved memory pages by XOR operation to get incrementdata of memory dirty pages. Then, compress the increment data which is easy to compress byXBRLE algorithm, and multicast the compressed data to target VMs. The target VM restoresthe memory data according to the information of incremental compressed data. Theincremental compressing technology can reduce the network overhead obviously indeployment process. DCVM increase VM deployment efficiency significantly while thesystem cost decrease in the process.