Abstract: In the cloud environment, the virtual machine deployment directly affects the overall performance of the data center. The concept of virtual machines affinity is proposed according to the relationship attributes among virtual machines. And a deployment strategy based on multi population genetic algorithm combined with the penalty function method is applied considering the load balancing of the physical machines and the affinity of the virtual machines. In order to avoid the local optimum, Gauss learning is carried out on the optimal individuals. Simulation results show that the deployment strategy with high load balancing and good affinity could be achieved through the multi population genetic algorithm, which has strong robustness, fast convergence speed and can solve the virtual machine deployment problem effectively in the cloud environment.

Key words: Virtual machine deployment, affinity, load balancing, penalty function method, multi population genetic algorithm, Gauss learning

摘要： 云环境中虚拟机部署问题直接影响数据中心的整体性能。针对虚拟机间亲和互斥关系提出虚拟机亲和度概念，综合考虑物理机负载均衡度，结合罚函数法思想，提出一种基于多种群遗传算法的优化部署策略。同时，为了避免陷入局部最优，对最优个体进行高斯学习。仿真结果表明，提出的多种群遗传算法获得了很好的物理机负载均衡度，可满足虚拟机之间存在亲和与互斥复杂关系时的部署要求，同时具有较强鲁棒性和较高收敛速度，能有效解决云环境中虚拟机部署问题。

关键词: 虚拟机部署, 亲和度, 负载均衡, 罚函数法, 多种群遗传算法, 高斯学习