The Application Of Virtualization In The Cluster Adaptive Power Management

Cloud Computing makes customers use computation and service allodially and pay money according to the usage for its innovative business model, this simplifies enterprises infrastructure construction and economizes the operating costs immensely. With the deeply development and application of the Cloud Computing, data centers play a more important role as the basic infrastructure of Cloud Computing. Nowadays, ,many big businesses invest heavily on the Cloud Computing project in order to keep competitive advantage, the construction of data centers increases vastly, but there exist many problems such as energy-hungry, low resource utilization and environmental pollution, which restrict the development of the data centers. The utilization of virtualization technology may be a good solution to these problems, now most research of data centers energy efficient is built on the virtualization, virtualized clusters have many advantages in energy efficient, you can manage and schedule computing resources effectively, adjust clusters running state properly, and then reduce energy consumption of the total system.The topic comes from National Natural Science Foundation program. The program hopes to build a new data center resource management system, which can allocate and schedule the data centers computing resources more rationalized and intelligently. To achieve this goal, we develop CREMS system based on open source cloud resource management software OpenNEbula.The paper's main contributions include:1) Build a new resource management system based on OpenNEbula, the system supervises all the physical and virtual machines of data centers uniformly, the administrator can get the CPU utilization,Memory utilization and QoS (Quality of service) in time, in order to manage and supervise virtualized clusters.2) Enhance the dynamism and rationality of the data centers resource allocation, this paper presents the system which can adjust the total resource of the data centers according to the information collected from all the virtual and physical machines. Each signal server can allocate resource to virtual dynamically according to its priority or performance requirement. As to the whole cluster, it uses workload consolidation to improve resource utilization, reduce energy consumption.3) CREMS system requires application level metrics like response time which is easy to find in web servers or database servers. CREMS will make a reasonable resource demand prediction according to response time and resource utility measured by CREMS, after predicting the workload of servers, CREMS will adjust the resource allocation of each virtual machine. When some of the physical servers are idle, CREMS will turn them into suspend state to save power, and will wake them up when necessary.4) Research on algorithm and policy of resource allocation. According to the information collected from virtual and physical machines, we design a high efficient schedule algorithm, allocate resource to each virtual dynamically, adjust the power consumption of physical machines , guarantee the stability and durability of the system.This paper verifies the relation between CPU utilization and power consumption, the result shows that the CPUs of cluster consume most of the energy, then we run the CREMS in the experiment platform based on OpenNEbula, Through adjusting workload dynamically, in condition of ensuring the QoS. The result shows that CREMS effectively reduced the total power consumption by about 12%.