Research On Clustering Performance Optimization In Cloud-environment

With the vigorous development of the Internet and the widespread of information technology, cloud computing technology brings human beings unprecedented network service experience. The issue how to set reasonable dispatch for the task cloud which submit to computing data centers and save energy during task execution has extraordinary significance for the development of technology.Cloud computing virtualization technology determines its host device and the underlying physical data centers can be made consisting of heterogeneous hardware environments. At the same time, the scale data centers have exponentially growing trend. It is a worthy research that how to distribute resources reasonably for the tasks in large-scale heterogeneous environments, and how to make the energy it takes to minimize during performing is also a hotspot.In this thesis, based on the research of the traditional scheduling algorithm, through the analysis of task scheduling strategy, the raising of the concept of task scheduling solution space and combining AFSA and thought tabu search, an improved AFSA(Improved Artificial Fish School Algorithm, IAFSA) was proposed. The algorithm reserves the AFSA total task execution time as a function of optimization and adopts the linear encoding that each N-dimensional vector represents a specific scheduling scheme. By using solution vector artificial fish directly as a method, the algorithm of artificial fish can be run directly in the solution space. TS for thought, not only retains the advantages of AFSA that converges fast during a large base computing, but also make full use of the advantages of Tabu search does not fall into local optima.Secondly, the thesis presents an improved Auto Scale algorithm for energy consumption optimization.(Improved Auto Scale Algorithm, IASA). IASA introduced the idea of the optimal frequency based on Auto Scale algorithm. Compared to Auto Scale,it has the advantages that maximize the amount of energy proportional tasks and set aside some extra capacity without increasing the energy consumption of the situation in order to deal with unexpected little load. Finally, the corresponding simulation experiments are designed for the above two task scheduling strategies. The experimental results show that twice scheduling strategy has a good performance compared to similar algorithms.