Optimization Of Power Management In Data Center Based On Elastic VM Pool

Nowadays, increasing importance has been attached to power management inmodern data centers. It has been a hotspot in computer science field. With the rapiddevelop of virtualization technology,it has been widely used in modern data centers.It improves data centers'reliability and ease of management, also contributes a lot inpower saving. But it also brings a lot of problems. The traditional power manage-ment solution cannot simply apply to data centers that used virtualization technology.Existing methods are excessive reliance on workload consolidation achieved by vir-tual machine live migration. Therefore, the study of reasonable resource allocationand effective power management is important for data centers that used virtualizationtechnology.The topic comes from National Natural Science Foundation program. The pro-gram hopes to build one effective power management framework for data centersthrough existing virtualization technology to improve resource utilization and powersaving. To achieve this goal, we develop elastic VM pool and build a power manage-ment framework based on that.The paper's main achievements include:1) Build power management framework based on elastic VM pool. This frame-work can collect all virtual machines'information including CPU and memory utiliza-tion in the whole data center. Resource that allocated to each virtual machine can belimited or adjusted.2) Improve the dynamicity and rationality of resource allocation based on elasticVM pool. Based on the information collected, the framework adjusts resource alloca-tion of whole data center. Each signal server can allocate resource to virtual machinedynamically according to its priority or performance requirement. As to whole cluster, it uses workload consolidation to improve resource utilization. Considering the ?uc-tuations of workload in data center, the framework uses two policy to achieve the goalof workload consolidation. In low-workload state, live migration of virtual machine isused while workload switching is used in high-workload state. In the mean while, wedevelop a method of fast deploy of virtual machine. It can create and boot a virtualmachine to provide service in a very short time. Power can be saved by this way withthe guaranty of service quality.3) Dynamic adjust the resource allocation and state of virtual machine, includeincrease and decrease of resource that allocates to virtual machines.The resource weadjusted is CPU and memory. Through the migration, suspend and resume of virtualmachines, the framework can control the power consumption of whole data center.4) Research on algorithm and policy of resource allocation. Based on the infor-mation, we embedded algorithms in our power management framework to improveresource utilization and power saving with the constraints of performance.We firstly verify the policy of fast deploy, result shows that it only needs 10% ofpower and time consumption with the compare of virtual machine clone. Then, by dis-tributing the framework on a simple data centers and running 1998 world cup trace asbenchmark, paper proved that the framework can save 5.5% power consumption withthe needs of performance. It shows that our framework can achieve the goal of man-aging resource allocation more reasonably and intelligently, and improving resourceutilization and power saving.