Research On Virtual Machine Migration Method Combined With The Measurement Of Performance Interference Between Virtual Machines

In recent years, VM live migration becomes a hotspot as an important branch in the field of cloud computing. It allocates the VMs dynamically to realize the server load average and ensure the quality of service of cloud application. On the one hand, virtualization technology solves the abstraction of physical resources, so one physical server can instance multiple virtual machines and deploy different application; On the other hand, because of virtual machines on the same physical machine compete for the same resources, it leads to the problems of performance interference between virtual machines. So this thesis, based on the in-depth analysis of the Xen virtual machine architecture, proposes a virtual machine migration methods combined with the measurement of performance interference between virtual machines.This thesis analyzes the research status of the virtual machine migration. According to the studies in virtual machine migration did not consider the performance interference between virtual machines, and proposes a virtual machine migration method combined with the performance interference between virtual machines. First, in order to solve the performance interference between virtual machines, this thesis uses principal component analysis and establishes a model of application's load to reflect the impact of application's performance. Then, based on this model, for virtual machine performance on the historical experience of performance interference, this thesis establishes a model to measure the performance interference between virtual machine, and estimates the parameter using multiple non-linear regression technique; For non-virtual machine performance on the historical experience of performance interference, this thesis proposes virtual machine load patterns mining based on Fuzzy c-means clustering, and gives a performance interference prediction algorithm between virtual machines reusing the history experience of virtual machine performance interference. On the basis of these models, this thesis proposes the selected algorithm of virtual machine to be migrated combined with the performance interference between virtual machines. In this algorithm, the goal is the virtual machine performance interference and the virtual machine migration cost. It can obtain a set of virtual machines to be migrated. According to placement of the virtual machine, this thesis proposes a virtual machine placement algorithm considering performance interference between the virtual machine. it bases on the model of virtual machine performance interference and finds a physical machine meeting the needs of the target virtual machine with the lowest degree of performance interference. Finally, based on the Xen virtualization platform and CloudSim cloud simulation platform, this thesis establishes an experimental environment and carries out a series of experiments. The result shows that this thesis's method can greatly reduce the number of virtual machines to be migrated while guaranteeing the performance of cloud applications compared with the existed methods of virtual machine migration.