Research On Power-aware And Performance- Guaranteed Virtual Machine Placement In The Cloud

With the development of virtualization technology,cloud service providers can provide ondemand,reliable and low-cost virtual machines(VMs)as services to users over Internet.Virtual machines in the cloud platform can ensure that cloud service providers provide users with computing,storage,network and other services via the Internet.Since VMs are running on the physical machines(PMs),the continuous expansion of the cloud platform will lead to an increase in the size of the PMs.Therefore,it is necessary to consider the great energy consumption of the PMs in the cloud platform.On the other hand,multiple VMs deployed on a single PM suffer from VM performance degradation due to resource sharing and contention.To solve the PM high energy consumption and the VM performance degradation problems in cloud platforms,it's essential to find an effective method to deploy VMs.The existing virtual machine placement(VMP)methods are mainly from the perspective of reducing energy consumption and avoiding SLA violations,and does not consider the performance degradation of VM,so the existing methods cannot guarantee VM performance and may affect the user experience.To solve the high energy consumption and VM performance degradation problems,this dissertation intends to study the VMP method from two aspects: the PM energy consumption and the VM performance.In this dissertation,a power-aware and performance-guaranteed VMP method(called as PPVMP)is proposed,which takes into account the PM energy consumption and the VM performance simultaneously.The main contributions are as follows.First,the dissertation constructs the PM energy consumption model.It measures the PM energy consumption data under different CPU utilization rates,excavates the relationship between power consumption and PM CPU utilization and then builds the energy consumption model.According to the experimental fitting results,the quadratic polynomial fitting model has the best result compared with the linear fitting model and piecewise linear model in relationship between the energy consumption and PM CPU utilization.The energy consumption model is helpful for the following VMP.Second,a method of VM performance model is proposed.As multiple VMs are deployed in a physical machine and share the physical resources.Therefore,the competition between VMs becomes more intense with the increase of the number of VMs,and may lead to the VM performance degradation.This dissertation measures the relative VM performance data under different PM CPU utilization rates,excavates the relationship between VM performance and PM CPU utilization rates and then builds the VM performance model.According to the experimental fitting results,the piecewise fitting model has better result than the quadratic polynomial fitting model in relationship between the VM performance and PM CPU utilization.The VM performance model is helpful for the following VMP.Third,the dissertation proposes a power-aware and performance-guaranteed VMP method.Based on the PM energy consumption model and VM performance model,we formulate VMP as a bi-objective optimization problem,which tries to minimize PM power consumption and guarantee VM performance.And the dissertation then proposes an algorithm which is based on ant colony optimization(ACO),to apply ACO in the PPVMP,we describe the definition of the pheromone trails and heuristic information,pheromone updating rules,and probabilistic decision rule.Finally,the dissertation uses the Cloud Sim as the simulation framework to evaluate the PPVMP with other VMP methods.The results show the efficiency of our algorithm.