中文摘要：

随着云计算和虚拟化技术的发展,为云资源管理提供了一种更高层次的调度选择：一个作业不再只能分配到一台物理机上,而是可将一台或多台物理机的计算资源虚拟化成一台虚拟机来运行该作业.根据作业需要,高效分配定量的物理资源放置虚拟机,是决定云系统性能的关键因素,即云资源调度问题实质就是一个虚拟机和物理机之间的映射问题.文中借鉴网络效用最大化模型,提出了一种云资源调度模型——云效用最大化（Cloud UtilityMaximization,CUM）模型,与传统调度模型相比,目标函数不再是最小化最大完工时间,而是以达到效用最大为调度目标,可以充分提高用户的满意程度.通过求解CUM优化问题得到最优的虚拟机和物理机映射关系.设计了针对该模型的分解优化算法——简化次梯度算法求解拉格朗日对偶问题,证明了该算法可以获得原始模型问题的最优解.仿真实验表明算法可行且具有良好的收敛特性,并给出了CUM模型在真实云环境下的应用场景.

英文摘要：

With development of cloud computing, especially wide application of virtual machines, it is possible that one or more physical machines can be virtualized one virtual machine to support a job. Virtual machine placement is a key factor for cloud performance, which is practically a mapping problem between virtual machine and physical machine. This paper gives a cloud schedu- ling model, Cloud Utility Maximization （CUM）, which applied approach of Network Utility Maximization into computing resource scheduling. Comparing with traditional scheduling prob- lem, CUM aims to maximize the utility of cloud instead of early finishing time. The optimal virtual machine placement policy can be achieved by solving CUM. The decomposition and opti- mization algorithm for the CUM, subgradient algorithm for solving Lagrangian relaxation dual problem, is proposed. Convergence of the algorithm is verified by the simulation experiments. At last the paper gives an application scenario of CUM in cloud environment.