Research On Virtual Machine Placement And Live Migration In Cloud Data Center

Cloud computing is a new commercial calculation model and service mode, and it distributes the computational tasks on resource pool which are consist of a large of computers, to make various of application systems get the computing power, memory space and information service according to the needs. At present, cloud computing has began the relevant applications and researches on web search, scientific computing, virtual environment and energy and biological information.With the development of cloud computing, the infrastructures are becoming increasingly large scale. Cloud data center can potentially house thousands of physical servers, and there are various services and applications running on them. Common in these systems is the resource sharing among a variety of applications with different resource requirements and possibly dynamic workloads. Virtualization offers a new approach to sharing resources, by allowing the provisioning and customizing of computing environments as needed and migrating workloads to adapt to changes. However, the growing management complexity due to the increased system size and the inherent dynamism experienced by the system pose great challenges in virtual machine resource management of cloud data center.Based on a comprehensive analysis of cloud computing technology and related works of data center resource management, this dissertation focuses on virtual machine (VM) placement and live migration of virtual machine in cloud data center, and gains several achievements on some sub-topics. The major contributions of this dissertation are as follows:1. This dissertation proposed multi-objective optimization for initial virtual machine placement. We firstly presented a management framework for virtual machine placement in an IaaS environment, then defined the initial VM placement problem as a multi-objective optimization problem and finally proposed multi-objective optimization for initial virtual machine placement based on Ant Colony Optimization (ACO). This algorithm is a distributed optimization method, which is beneficial to parallel computing. It has the positive feedback mechanism, and through the pheromone is constantly updated, it can get the optimal solution by the efficient convergence. Experimental results show that compared to heuristic method and genetic algorithm, our algorithm can achieve the optimal balance in multiple conflict objectives, which effectively reduces the resource wastage and power consumption, and minimize violation of SLA in datacenter.2. This dissertation proposed a multi-objective approach for dynamic virtual machine placement. The system considered three conditions for dynamic VM placement which included of improving the quality of service, reducing the resource contention and decreasing the power consumption. The management approach concentrated on the dynamic VM placement problem due to changes of system conditions or the application workload, i.e. deciding when to migrate VM, which VM to be chosen and where to place VM. On the problem of when to migrate VM, we set high and low threshold for different optimization targets and used the sliding-window and time-series prediction techniques to judge whether triggered the virtual machine migration. On the problem of which VM to be chosen, the system adopted different tactics to choose the virtual machine according to different optimization targets. On the problem of where to place VM, the system chose the host node by the TOPSIS method which was to solve the multi-criteria decision analysis method, and the purpose was to balance the conflict between multiple targets in the system. Experimental results show that the multi-objective approach based on TOPSIS can get less violation of SLA, resource load and power consumption, and balance the conflict between different objectives.3. This dissertation proposed a fast live virtual machine migration method. Through the formal definition and the performance analysis of the pre-copy model, a bitmap page which marked those frequently updated pages is firstly added in the pre-copy approach. Then we used the time-series method to judge the memory pages the iteration process. The dirty memory page in this iteration whose dirty times in the previous iteration were more than the given number would be marked the frequently updated page. Those marked pages are transmitted in the last round of the iteration process only, thus avoiding the memory pages of repeated retransmission. A contrast between the pre-copy approach and the fast pre-copy approach is given. The fast pre-copy approach can distinctly reduce the total migration time and transferred data of live migration, and achieve the purpose of fast live virtual machine migration.