| As cloud computing technology becoming mature and popular, system virtualmachine (VM) technology is more and more being applied in the information system.How to deploy the VM to each node quickly and make the VM run efficiently is one ofthe research hot points nowadays. In the present cloud platforms, to deploy the VM byVM image solves the batch deployment issue of the operation system and software, butas a result, the operating system and application software will be coupled together bythe VM image, which leads to the following disadvantages in the VM deployment underthe cloud platform:1) the distribution is slow in the VM deployment because of therelatively large size of the single VM image;2) Users different needs for computingenvironment lead to the VM image redundancy, low resources utilization, storage spacewaste and trivial VM image management;3) The software management operationresults in the redeployment of the VM, which normally leads to the unstable servicedirectly.Aiming at the problems in the existing cloud platform, such as relatively large sizeof the VM distribution image, VM image redundancy, low resource utilization rate,unstable service, etc., and then after referencing the latest software deploymentmechanism, such as APP store, this thesis designs an on-demand deployment operationsystem for the operation system--OpenAppV based on the VM technology. Throughseparating the application software from the VM image and then publishing them on thecloud platform as services, users can choose the OS and application software accordingto their own demand to construct user-defined VM system.First of all, according to the characteristics of the VM software distribution, thisthesis puts forward an on-demand software deployment mechanism which is based onthe pre-installation. This mechanism adopts the static installation way in the VMstarting process to deploy the software to the VM system. Through the analyzing ofsoftware packages in different formats and the resolving the description files of thedependency relationship, all the description files of the dependency relationship areunified in one single format. Based on the features of the software dependencyrelationship, adopting the dependency deployment of the parallel work, the timespending on the software deployment to the VM image is reduced in paralleldeployment mode. To check if the target system environment is proper for the softwareexecution, a software deployment judgment strategy is brought forward. So, the targetenvironment can be tested and adjusted in both software deployment constraint conflictdetection and software deployment de-dependency conflictThen, aiming at improving the speed of the software distribution, according to thefeatures of the cloud platform, this thesis introduces and designs distributed multi-level cache model based on the regions, and based on this model, puts forward the softwarepackage prefetch strategy based on the sharing subset for multiple-user access to thesame software in the same region; meanwhile, proposes the software package prefetchstrategy based on the access frequency and object correlation for the frequently accessedsoftware.Finally, on the open source cloud platform OpenStack, the prototype of OpenAppVis implemented. The function and performance of the prototype are tested. The resultsshow the feasibility that software can be combined on-demand through the deploymentand VM image. Also, the VM distribution is speeded up when the VM is beingdeployed. This effectively resolves the VM image redundancy problem, saves thestorage space and simplifies the VM image management. Eventually a fairly goodbalance between the system performance and resource utilization is achieved. |
PDF Full Text Request