Research On Key Techniques Of Energy-Efficiency Modeling And Optimization For Cloud Computing

Diversified demands for users and service providers require higher service quality and system performance with cloud computing technology. High energy efficiency cloud with high performance, low energy consumption and QoS guaranteed is one of the most challenging topics and hotspots for cloud computing. This dissertation discuss the key technologies of high energy efficiency cloud computing, the main contents and relevant achievements are as follows.(1) The current researches focus on evaluating and optimizing related performance metrics of QoS without considering quantitative analysis between QoS and the energy consumption of the system. By constructing the throughput per energy consumption model and introducing this energy efficiency model into QoS, the QoS evaluating mechanism based on energy-efficiency-driven is proposed. By analyzing and calculating the energy efficiency value of the computing nodes, the existence and occurrence conditions of the energy efficiency maximum are deduced, and an energy efficiency measurement model of the cloud is built. By studying the trade-off relationship between the energy efficiency and the metric parameters in QoS, which is analyzed by quantitative approach of M/M/n query theory, an energy-efficiency-based QoS evaluation method is presented. Based on the proposed energy efficiency valuation model and energy-efficiency-driven QoS evaluating method, an optimized resource allocation strategy for different service requirements is designed by using the evolutionary game theory.(2) High Energy efficiency is one of the most important issues for large scale server systems in current cloud computing. The main method about the trade off is minimizing or maximizing one factor with fixing others. An effective integration energy-efficiency evaluation model is needed, which can describe the cloud "degree" of energy efficiency better. Then, QoS parameters reduced approach and a weighted energy efficiency model is investigated, which introducing the performance indicators into QoS. Reducing the discrete QoS parameters to a unified dimension, thus the energy efficiency can be calculated through the QoS/energy consumption. In addition, according to the different energy efficiency the cloud datacenters is divided into different energy efficiency classes and levels. The qualitative evaluation about the energy efficiency is implemented.(3) The reasonable degree of resource allocation is the key factor influencing the cloud computing energy efficiency. Existing scheduling mechanisms allocate resources and schedule jobs by priority-based strategy come up somewhat short. Fixed resource allocation granularity lead to inefficiency. Such mechanisms have little awareness of the changes of cluster running status and job execution status. It is difficult to mapping the enengy efficiency to a certain resource allocation. Based on that, an Energy-efficiency-aware Scheduling method of virtual resource allocation is addressed. Using the granular computing and the relevant theories, an energy-efficiency-aware-based and granularity space-seeking optimization-based method is proposed. Choose a suitable task granularity and resource granularity with energy-efficiency-driven. Co versioning the granularity hierarchy according to customer demands. Guarantee the user's QoS by multi -granularity. Based on the proposed method can make the cloud environment staying high energy-efficiency.(4) The main method about the power-performance-QoS tradeoff by fixing some factors and minimizing or maximizing the other still exist several main challenges: Involved varied nenergy efficiency parameters combine close coupling. It's hard to build its accurate math model. The deviation of building model and the influence of interference factors can lead the uncertain results of energy efficiency value and disrupt the stability without the robustness. Multivariable fuzzy decoupling control converts MIMO system into several non-interfering SISO model without building exact mathematical models, and maintain the desired performance in spite of the disturbance to system with specified uncertain bound. Adopting the fuzzy control theory, the fuzzy rule and fuzzy membership functions about energy-efficiency is set. Both FNN (Fuzzy decoupling device) and Fuzzy Predictive Decoupling Controller is designed, a robust energy-saving technique and a high energy-efficient model is built.