A Study On The Policy Of Placement Strategy Of Virtual Machines Based On Workload Characteristics

Virtualization technology abstracts the physical resource layer for centralized management and scheduling, which can improve resource utilization. And now virtualization technology is widely used in server systems and data center technology, to improve the resources utilization and provide the services on-demand. When there are several virtual machines (VM), the computer systems will over-provide resources to meet the customers’ needs, which result in the low utilization and a large number of idle resources. Before the virtualized environment providing services, virtualization machines reasonable placement will save costs and the good resource providing and Scalability on the virtualized environment. To this problem, this paper researched the VM placement strategy based on the workload characteristics.About the traditional method of VM placement methods, there mainly are the peak distributions of placement methods, the average distribution of placement methods. Those methods apply for all virtual machines, but the over-providing resource will waste the resource. For the complemetarity of the workload, there are new approaches like the one-time matching peak distribution of placement methods and the one-time matching average distribution of placement method and so on. Complementary is divided into time and space complementary. The time complementary is the complementarity of the same resource at the same time in different VMs and the place complementary is the complementarity of different resources at the same time in different VM. This approach improves the loss of idle resources and the users’service quality caused by the peak or average allocation.In this paper, we research the VMs’workload characteristics, model the CPU utilization of different VMs as VMs’workload characteristics and propose the virtual machines placement strategy based on workload characteristics. This strategy preprocesses the VMs’ queue and shields the high-work VMs and calculates the complementary of different VMs. In order to control the iterative matching times, we propose to select the two machines which has the strongest complementary and add the workload of the two machines, and then test the saturation. Through controlling the saturation, we decide to put the sum of the two machines into the VMs queue and do the iterating matching operation, which can achieve the aim of matching multiple VMs and utilize the idle resources.In this paper, in order to verify the strategy’s reasonableness and effectiveness, we did a lot of experiments using the automated evaluation tools httperf and autobench on the XEN virtualization platform. Through different matching strategies, we compare the results to verify the strategy we proposed is reasonable and makes the VM to meet the user quality of service while improves the utilization of physical resources. Those experiments show that the proposed method can improve resource utilization by37.5%than the traditional method, by12.5%compared with non-iterative matching method. With this approach, we use fewer physical machines while maintain essentially the same VM performance.