The Resource Provisioning Mechanism Based On Energy Efficiency And Fairness In Cloud Computing Envinronment

As an emerging commercial calculation model and service pattern(IT as a Service), Cloud Computing techniques have been becoming the research hotspots attracting extensive attention in information field. With the help of virtualization, cloud computing system distributes computation tasks finished through the resource pool that is made up of a large amount of computers, and in on-demand manner dynamicly provides elastic scalablibility services to Internet users(computation services, storage services and information services). Today, Cloud Computing has been forwarding the corresponding application and exploration in the fields of internet search, scientific calculation, virtual environment, energy, biological information and so on.Along with the fast development of cloud computing, the scale of cloud computing datacenter infrastructure is sharply becoming larger. However, with the expansion of the system, which is becoming more and more complex. High performance computing will develop along with the direction of high energy utility, not be from single pursuit to high performance of system. National medium and long-term strategic planning also has cleared the target of the energy-efficient computing in the development of IT in future. Energy consumption, as a system requirement, on the one hand, is influenced by resource execution efficiency. On the other hand, which also influences the users in the form of resources or service cost. As the technology innovation, Server's initial acquisition cost is tending to decline, however, it is continuing to rise that capital spending, operating expenses and the environmental impact of energy consumption brings out related costs. Therefore, it is becoming the development trend of the technology of computer and IT industry to improve the efficiency of power consumption so as to achieve green computing.The productivity of system energy efficiency not only depends on the system hardware (such as system architecture and manufacturing technique), to a large extent, but also depends on the resource management system deployed on the architecture. Resource management is quickly responsible for the response to a user's resource requests, the fair effective scheduling tasks and the reasonable on-demand allocation of resources, so as to ensure that users expect performance and efficiency of the use of system resources. The main task of resource management in virtualized datacenters:dynamic resource management, elastic responding for user requirements, on-demand automation resource allocation, energy saving for infrastructure have been becoming the approbatory development trend of the industry. Energy saving oriented adaptive resource management is one of the hotspots of current research on green virtual datacenter. In virtualization datacenter, Virtual Machine (VM) is the basic unit of resource management for both user requirements and task scheduling. Therefore, VM management(such as deployment, migration, revocation and so on) becomes to the core of resource management. The multi-VM scheduling problem, in consideration of its NP complexity, Resource heterogeneity, the compromise between the new user requirements of applications and the scheduling goal, and so on, has not been very good to solve at all, especially under the background of energy consumption that has become the important factors of influencing global energy and the environment. In addition, From the global perspective, load imbalance of the whole system resources also can lead to system operation cost and lower efficiency. The related work of the thesis is based on these. First of all, analyzes and summarizes the major resource management adaptive methods in the current data center and the research situation on resource provison of IaaS, and then several key problems around resource energy efficiency in virtual data center resource management were studied. The main work includes as follows:1) The thesis proposed a strategy of energy-aware virtual machine deploymentThe bidirectional heuristic greedy algorithm search based on CPU and the Memory searches for resource in the direction of the two key computing resources, ensures load balancing during virtual machines deploying, based on the virtual machine live migration and consolidation under the double thresholds, which reduce service default rate and energy consumption. At the same time, proving that system total energy consumption is minimized when load balancing and verifying the effectiveness of the algorithm implemented. By adopting the dynamic weighted idea based on different service types resource prediction, forecasting more accurate. Adopting the method of quadratic exponential smoothing predicting that the local server resource requirements, which reduces the secondary migration after the virtual machine deployment, enhances the stability of the system and saves energy consumption. Meanwhile, adopting dynamic weighted idea based on time-unit, predicting power value in the next cycle is based on the predicted and observed values of the past cycle which makes error weight minimum, which effectively eliminates the accumulation of the prediction error of predicting distortion and increases the load forecasting accuracy of system. Experiments show that the strategy in load balance, energy consumption, the virtual machine migration have very big promotion comparing with the algorithm based on the CPU energy awareness. 2) The thesis proposed a strategy of spatial/temporal-aware virtual machine deploymentMulti-objective maximum energy efficiency optimization algorithm based on maximum time sharing of the VMs deployed on the same physical machine. Ensuring load balancing of multiple resources during the much longer time in the phase of VM deploying, but also reducing the potential migration of virtual machines and the performance disturbance brought out by VM migration, and energy consumption, enhancing the stability of the system. Dynamic power on/off server strategy without losing performance at the same time, further reduce the energy consumption of the system. Based on the average resource utility of current system and time overhead on the server powering on/off, adaptively decides the time of servers booting or shutting, further reduce the system energy consumption without losing performance at the same time. The experimental results also verify the superiority of the strategy.3) The thesis propsed a strategy of multiple resource fairness allocationBased on user request dominant resource entropy and dominant resource weight, the thesis proposed a multiple-resource-joint fair resource distribution. Adopting the strategy for virtual machine deployment, to achieve Pareto optimal allocation of resources and prove the corresponding property. Dominant resource entropy guarantees adaptability of VM deployment to server resources, dominant resources weight and Max-Min strategy ensure the order of VM competitive advantage resources. In addition, On the premise of guarantee resource allocation fairness, improves the utilization rate of the datacenter resources so as to improve energy efficiency in the data center. Experimental results based on Google's cluster usage data set clearly showed the performance of the proposed algorithm.The thesis carries on the deeper research around the three sides of load balancing, energy consumption and fairness for resources provisioning in cloud computing data center. The thesis proposed three algorithms which are the bidirectional search heuristic algorithm based on the CPU and Memory, the energy efficiency maximization algorithm based on virtual machine service time and space, and multiple resource fair allocation algorithm based on dominant resource. Which can provide reference and supporting for the following study of supply mechanism in cloud computing data center.