| As a kind of novel information processing model, cloud computing technology has become a focus in the field of information. With virtualization as a support technology, cloud computing provides a kind of dynamic and scalable service for Internet users on demand. However, as cloud computing environment is characterized by the large-scale, the dynamic and flexibility of resource management and allocation, the energy comsumption problem and resource provision efficiency are critical factors that impact on the performance of cloud computing. Based on the new computing infrastructure—cloud computing technology, this thesis focuses on how to optimize the energy comsumption and the resource allocation of cloud data center. So far, there are still many problems to be solved concerned the energy consumption problem and resource provision of cloud computing. This thesis focuses on the resource provision optimization methods of cloud computing environment from three aspects:energy-saving mechanism, load balance and market economy model. The main contributions of this thesis include:1) Study systematically on energy-saving mechanism and optimization method of resource provision in cloud computing environment.First, starting from the basic concept of cloud computing, the characteristic, the service type and layer of cloud computing are introduced respectively. Second, the energy-saving optimization strategies in cloud computing are stressed, the application environment and the advantages and disadvantages of them are analyzed and compared as well. Then, the resource provision technology is studied further and the optimization strategies of this field are classified and compared. At last, the experimental tool CloudSim is introduced and the mechanism of its resource provision is analyzed through several simulation experiments.2) A virtual machine resource provision strategy based on the equilibrium between energy and SLA is proposed.Aiming at dynamical changes of application workload requirements, a self-adaptive deployment strategy RLWR based on robust local weight regression is presented. The algorithm can decide the overload time of hosts dynamically according to the historical resource occupation information of application workload. After detecting overloaded hosts, two virtual machine migration selection algorithms, MPM and MNM are proposed. The objective of the former is to get optimal migration period, the later is to get minimal migration number. The migrated virtual machines are deployed using bin-packing algorithm PBFDH based on power consumption aware. Contrasting to static threshold algorithm STH, MPA and DVFS, virtual machines are not only deployed on fewer physical hosts in the self-adaptive deployment strategy, which promotes energy efficiency through turning off unused hosts, but also the load prediction of resource can bring high-reliable QoS delivery and avoid overmuch SLA violations between users and resource providers. The experimental results show that the strategy has an obvious effect on decreasing SLA violation under ensuring energy-efficiency.3) A green energy-efficient resource provision strategy based on multiple data centers is presented.The energy efficiency of data center is usnally affected on many dynamic factors, including energy costs, carbon emission rate, load type, CPU energy-efficiency and cooling system and so on. The above factors are considered in the proposed strategy and the global energy-efficiency of data center across multiple geopraphically heterogeneous environments is studied. First, the resource provision model of multiple data centers is set up, the profit and carbon emission issure influenced by energy comsumption is formatted as a multi-objective optimization model of the profit function and the cost function with QoS constraints. This model is proved to be an NP-hard problem. Aiming at this issue, four algorithms are proposed, CMM, MCMP based on green cloud priority and PMM, MPMC based on profit priority. The carbon emission, energy cosumption, profit and QoS requirements are considered in algorithms synthetically. The objective is to reduce carbon emission and increase revenue with meeting QoS requirements of user applications. During implementing applications, for optimizing energy consumption further, the optimal frequency of CPU is sovled and we prove that the energy consumption will reach a local munimum at this frequency. The experimental results show that the strategies not only reduce the energy costs, otpimize the task scheduling and also balance the carbon footprint.4) A virtual machine resource provision load balance strategy based on genetic algorithm is proposed.The diversity of application demand and the heterogeneity of resources inevitably lead to load imbalance of cloud computing during the process of resource provision, which greatly reduces the global efficiency of resource provision. How to improve resource utilization and system performance through an efficient load balance mechanism is an urgent need to solve currently. Aiming at this problem, a virtual machine resource provision genetic algorithm based on load balance, VMPGALB is presented. Abandoning the traditional binary encoding, VMPGALB adopts a tree encoding scheme, which is more suitable to reflect the characteristics of virtual machine provision. During designing the selection strategy, the proportion selection based on fitness and the elite-preserving strategy are applied, which make individuals with smaller fitness to be selected possibly and retain the best individuals directly to offspring. Druing devising the crossover operator, VMPGALB has a better crossover performance through the crossover operation of two parent individuals and applying spanning tree method. Meanwhile, in order to avoid a local optimum, VMPGALB implements mutation operations on individuals according to a certain percentage. The experimental results show that, contrasting to BGA, MOGA, BFH and WLC, VMPGALB not noly can get better genetic performance and fewer migration numbers, but also gain the load balance scheme of virtual machine provision with a faster convergence speed.5) A resource provision game strategy based on market economic model is presented.Market economics model can achieve the optimal allocation of resource through the equilibrium theory, cloud computing reource provision mechanism based on the market economic model is researched. Combined with the advantages of game theory in resource management, first, the resource provision model in non-cooperative competitive market is set up and a non-cooperative game resource provision algorithm RPANCG is proposed, which is based on non-cooperative game. The objective of RPANCG is to find Nash equilibrium that makes the utility of all resource providers optimal. We prove that RPANCG can generate a unique Nash equilibrium. Then, based on RPANCG meeting mutual optimum of the revenue, for increasing collective revenue further and meeting the efficiency and fairness, a resource provision algorithm RPABG is proposed in bargaining market based on non-cooperative competitive market. RPABG builds its model with bargaining game theory and its objective is to find Nash bargaining solution. The experimental results show that RPANCG can converge to a unique Nash equilibrium, the utilities of resource providers reach mutual optimum and the whole resource provision tends to be more reasonable. Moreover, RPABG can further give consideration to the efficiency and fairness of resource allocation based on RPANCG, improve the overall utility of resource providers and realize the Pareto efficiency, which leads to an optimal allocation of cloud resource with fairness, rationality and equilibrium.This thesis is supported by National Natural Science Foundation of China (No.60970064, No.61272116), New Century Excellent Talent Support Plan (No.NCET-08-0806), Specialized Research Fund for the Doctoral Program of Higher Educaion of China (No.20120143110014) and Hubei province high-end talent lead cultivation project. |
PDF Full Text Request