Low-cost Placement Of Virtual Machines In Cloud Computing

Cloud computing has developed rapidly in the past decades. As the number of users growth, the size of data center is become bigger, the cloud service provider(CSP)'s operating costs also increased dramatically. Studies have found that power consumption take up about 60% of the total operating costs, and network equipments and bandwidth consumption take up about 15%. How to reduce the power and bandwidth consumption and keep the quality of customer service has become a crucial issue.The development of virtualization technology allows a physical machine(PM) to run multiple independent virtual machines(VM). In cloud computing, CSP provide users the VM as the basic unit of cloud resources. We try to place the virtual machine with big traffics together to reduce bandwidth consumption of network and the requirements of the core equipment's performance, so it can improve the scalability of data center. We also can get the VMs closed to reduce the number of PMs used, it help to reduce the power of data centers.In this paper, we have done the research on static VM placement and dynamic VM placement as follows:In static VM placement, we assume the PMs and VMs all are heterogeneous and studied how to reduce power and bandwidth consumption in the data center. The problem is due to NP-Hard, it can't be solved in polynomial time, so we proposed the Improved Group Genetic Algorithm(IGGA). Our idea is hoped to optimize the global goals through each perfect combination of VMs in a single PM. When PM's inner-traffic is bigger, the external-traffic is smaller; and the less resources waste of single PM, the less number of PMs used, then the total energy consumption is smaller. Based on the idea, we improved the traditional Group Genetic Algorithm(GGA) and proposed a new algorithm to place VMs and improved the mutation operator. Through comparison testing in MATLAB, IGGA can reduce the power by 2.2 percent and bandwidth consumption by 4.3 percent relative to GGA.In dynamic VM placement, PM's resource utilization rate is changing all the time, CSP need to balance the load of each PM. We study how to choose migration VMs and find the destination PMs to reduce the power, migration times and SLA default rates in data centers. When choosing the VM to be migrated, we consider the current utilization of the overload PMs in each dimensions and corresponding threshold in cloud environment, and give dynamic weights for different dimensions in each overload PM; then we defined the perfect migration VM as idleR according to each VM's utilization in overload PM. Finally, we choose the VM which closest idleR in weighted Euclidean-distance out of the overload PM. O n the destination PM selection, we considered three factors: the PM's closed statues, the increased energy after migration, and the available resources. According to these three priority, we choose the best one as the destination PM. Finally, compared to other algorithms, our method can reduce more than 60 percent of the number of migrations and 20 percent of the SLA defaults through CloudSim simulator.