A kind of strategy for the deployment and scheduling of virtual machines is proposed based on multiple attributes analysis to solve the uneven loads problem among physical servers in the cloud computing. The strategy classifies virtual machines according to the characteristic of resources, and it is composed of two aspects. Resources of hot spots are analyzed to quantify their important degrees in the virtual machines deployment, and evaluate all the physical servers according to the virtual machine vector, then the best physical server is selected to deploy. The vectors of virtual machines which are running on the overload physical servers are obtained in the virtual machines scheduling, and the rest of physical servers are evaluated in the right order. This strategy not only realizes the optimal allocation of the resources and reduces the overall loss caused by dynamic load balancing. The experimental results show that, when 20 virtual machines in five physical machines are applied in the same order, the average number of dynamic migration of the proposed algorithm significantly reduces about 80% than that the random equalization strategy does, and the rates of physical server resources usage are more balanced.%针对云计算中物理服务器间的负载不均问题,提出一种基于多属性层次分析的虚拟机部署与调度策略.该算法将虚拟机按照资源的需求特点进行分类,主要由两方面构成:在虚拟机部署时,对虚拟机的资源进行热点分析并对其重要程度进行量化,根据量化后的权向量以及服务器资源的使用记录对各个服务器进行预测评价,选择最佳服务器进行部署；在虚拟机调度时,获得运行在超载服务器上的各个虚拟机的权向量,并按照一定次序对未超载服务器进行评价,查找是否有更适合的服务器,从而降低超负荷服务器的负载.与同类算法相比,该算法不仅实现了服务器各项资源的优化配置,而且降低了动态负载平衡导致的整体损耗.实验结果表明,当按同一次序在5台物理服务器上申请20台资源需求不等的虚拟机时,该算法到达平衡状态需要的平均动态迁移次数比随机均衡算法减少了80％,同时进入平衡状态后,各服务器的各项资源使用情况也更趋于平衡.
展开▼
