A Study On Adaptive Allocation Strategy And Mechanism Of Virtual Resource On Virtual Machine Platform

Virtualization technology is an abstraction of computing resources, which separating the hardware and applications, refining and automating the system configuration through data center, monitoring and management, changed the manual configuration of traditional management of computer resources, reducing the error rate, improving the efficiency and reducing the costs. It is able to quickly re-examining the server's resource requirements for each virtual machine to optimize resource sharing, thus to meet the application requirements, and give the more effective response for the changing things, which improved the performance of fault isolation, enhanced the independence of customer VM (Virtual Machine) and the separation performance. With the re-emergence and extensive application of virtualization technology, in virtual computing environment, the resource allocation of virtual machine management problem, namely, how to dynamic allocate the resource to meet the QoS (quality of service) objectives of application load, improve the utilization of system resource, reduce management complexity and the number of using physical machines, save resources based on the changes of workloads and resource requirements has become a research hotspot.In this paper, the virtual machine resource allocation management problem is studied, and the virtualization technology, Xen virtual machine and the forecasting methods in resource management are introduced in detail. What's more, summarization is made on dynamic adaptive resource allocation strategies and multi-machine resource management studies. On this basis, a dynamic resource allocation manager on a single virtual machine is established, as well as multi-machine resource management strategies and related algorithms are also proposed. The main contents of this paper are as follows:(1) Towards to the problems of resource management on single virtual machine, an adaptive dynamic resource allocation management mechanism is proposed, the resources allocation management control model is also established. The model is consisted of controller, distributor arbiter and the related allocation rules. In order to improve the performance of virtual machine and control the allocation of virtualized resources online effectively, this paper uses dynamic resource allocation techniques dynamically allocate resources to various virtual machines as the workload varies. Gray prediction model is used to predict the resource allocation value of virtual machine. This paper also adjusts the boundary conditions of grey prediction model to make the prediction more accurately. At the same time, the control theory is used to feedback control resource utilization to obtain desired resource utilization levels by regulating the value of allocation of virtualized resources automatically.(2) For resource management of multiple physical machines, this paper presents d如果ferent resource management models and algorithms according to d如果ferent application environments, and then designs resources allocation manager based on d如果ferent constraint. It develops relevant migration strategy for the virtual machine migration based on migration cost, resource utilization, load balance and the number of machines, and divides the resource management model into two phases. For the resource management of limited, QoS guarantees, certain utilization and load balance are main objective, Sub-Virtual Machine Set (SVMS) and Migration algorithms are proposed. The resource management is divided into local control and overall control phases. This paper also carries out theory proof of the optimal for SVMS algorithm, ensures load balance and resource utilization, meets the service quality requirements of applications.(3) For resource allocation management of no limit, the initial phase determines which VM should be migrated based on meeting QoS, maximizing resource utilization and minimizing migration overhead. It also presents the Search Virtual Machine (SVM) algorithm for this phase. The second stage formulates migration program, and then designs Improved Migration (IM) algorithm to migrate VMs, with the aim of least number of physical machines and the largest resource utilization, which reduce the usage of machine, ensuring the certain resource utilization.The CPU allocation control experiment on Xen is carried out for the single resource allocation management. The results show that when the target utilization rate changes over time, the model is able to track these changes quickly, so that the resources allocation value also varies with the change. The allocation value can achieve the target utilization requirements, and utilizations can quickly converge to the target value. The simulative emulation of multi-machine resource management shows that the SVMSM algorithm gets great improvements in resource utilization, migration overhead and load balance, and the SVMIM algorithm has good results in resource utilization, usage of physical machines and energy consumption of system.