Research On Energy-Saving Resource Scheduling Strategies In Cloud Datacenters

With the rapid development of cloud computing, the scale of construction about large data centers become more and more massive. Problems like low utilization of resources and high energy consumption followed, then efficient resource allocation strategies are developed. Research on energy saving technology of data centers in cloud computing environment has become a research direction. As an index of resource scheduling, energy consumption should be affected by server performance at first, then application requirements. Now virtualization technology mainly consisting of resource virtualization and reallocation is widely used in resource scheduling. The problem of virtual machine reallocation has proved to be NP-hard problem, together with the requirements of various kinds of applications and diversity of physical environment, energy consumption calculation has become challenging. Especially in the context of global energy shortage and the greenhouse effect increasing yearly, energy-saving scheduling strategy has become a subject worthy of exploring further.This thesis discuss energy consumption problem from two aspects of resource monitor system and energy consumption modeling. Then according to the engineering experiment server in the power consumption of different CPU utilization value, calculate the energy consumption of the parameters of the model. On this basis, this thesis study the criteria of virtual machine migration and the strategy of virtual machine selection for the server to be migrated. Based on the experimental data, some parameters are obtained. According to the comparsion results, which one is more able to adapt to this kind of small cloud laboratory datacenter, the MM strategy can achieve best effect on reducing energy consumption among three selection strategies.With the goal of load-balancing in DC, optimal target function is set up for solving VM allocation problem, yielding solutions close to optimal, while the average utilization rate of all physical servers is maintained at a desired level. In order to solve the optimization model, the corresponding genetic algorithm have been designed.Its chromosome encoding adopts the binary code. And the random selection,two point crossover, elite principle and other methods are used for the implementation. Then with the engineering projects that is often used in the packing algorithm and random placement algorithm to product the placement sequence,and then to get the fitness function value com-parison. The experimental results show that the server-scale CPU utilization calculated by genetic algorithm is the closest to the expected optimal value solved by optimal allocation strategy. Load balancing means effective reduce the number of server. Thus energy reduction is obtained.Finally, we systemically study energy-efficient schduling mechansim for the cloud system. A performance and energy consumption evalution model for the cloud system is proposed, and a performance-energy combined optimization model represented by a pure profit function is further established. Then, a genetic algorithm is developed to search optiaml solutions of the proposed optimization model. These optiaml solutions represented by request scheduling strategies and resource management strategies potentially imply the balance of performance and energy consumption, and thus are more reasonable than some optimal strategies for optimizing performance or energy consumption separetly. Numerial exmaples illustrate the analysis and evaluation method of the theoretical model, the realiztion of the energy-efficnet scheduling mechanism, and the significant effect can be achieved on the pure profit optimization.