Memory Optimization Technique Of Multiple Virtual Machines For Multi-objectives

With the rapid development of computer technology and the widespread use of computer application projects, cloud computing gets a high degree of attention in scientific computing and commercial fields as an innovative computing schema. Services offered by large data centers and virtualization technology make up the infrastructure of cloud computing. Memory virtualization is an important part of virtualization technology. The dynamic changes of application service concurrent requests and various applications have brought new challenges to the allocation strategy optimization of underlying virtualized memory resource. Multiple virtual machines in the virtualization system are short of efficient dynamic memory adjustment mechanism. Memory resource demand of these multiple virtual machines changes constantly, making the memory resource allocation unbalanced, such as large amount of memory resource in some virtual machines is idle, while some virtual machines are serious shortage of memory resource and so on. These phenomena result in poor service performance of all virtual machines and lower user experience. Facing complex services and applications, how to improve the global performance of multiple virtual machines and service quality is a research emphasis in multiple virtual machines'memory optimization problem.Based on the traditional methods of virtual machines'memory optimization, the memory optimization technique of multiple virtual machines for multi-objectives is proposed in this thesis. By analyzing the shortcomings of traditional memory optimization methods, multi-objectives of multiple virtual machines are optimized globally not just for local optimum in this thesis. Then the process of this memory optimization technology is given. The memory optimization technology mainly includes two key algorithms, one is virtual machines memory demand prediction algorithm based on RBF-Markov and the other is multi-objectives memory optimization method based on genetic algorithm. Firstly virtual machines memory demand is predicted through using the RBF neural network, and modified by using Markov model in order to improve its accuracy. Secondly the multi-objectives optimization model is built and solved by using genetic algorithm, and the balloon driving mechanism is adopted to adjust memory resource distribution among virtual machines, and then the comprehensive performance of virtual machines is evaluated by radar chart.The results of the experiments show that the prediction algorithm of virtual machine memory demand has high accuracy, and the memory optimization technique of multiple virtual machines for multi-objectives is feasible and effective. Through controlling virtual machine memory dynamically, the utilization rate of memory resource is improved while users are satisfied with the quality of service received. The radar chart evaluation results show that the performance disparities among virtual machines reduce and the comprehensive performance is improved.