云计算数据中心应用感知的动态资源配置技术研究

For the multi-mode applications, the switch of functional modes will lead to substantial difference on the resource requirement, since the type (mode) of application determines the characteristics of the underlying resource consumption. Hence, it will be difficult to achieve "on-demand service" if the cloud data center still conducts resource provision for the applications with a single schema. This project founds on the improvement of cloud data center resource utilization, takes the establishment of multi-mode applications oriented efficient resource allocation mechanism as the core content, starts from the analysis and characterization of the fine-grained resource requirement features for the multi-mode applications when running on the virtualization-based cloud system, we will construct the resource probability requirement topology based on the basic resource requirement topologies and the incremental resource requirements of different modes. Based on the association of incremental resource requirements between multiple applications, we will study the construction of application community that shares the incremental resource and the complex algorithm of resource requirements, and implement the resource configuration mechanism while taking computing cluster as basic unit. For the issue of virtual machine placement in the computing cluster, on the basis of the performance interference model between different types of applications, we will investigate the virtual machine placement strategy with balanced resource utilization. For the issue of the evolution of running application and the global optimization, we will study the dynamic resource adjustment mechanism for benefit maximization, while keeping system efficiency and stability. The above mentioned research outcome will be evaluated on the cloud system that developed by the project research group independently.

应用类型决定了其对底层资源的耗用特征，而复杂应用运行中功能模式的切换对系统资源的需求则会发生本质变化，如果云计算数据中心仍采用单一资源配置方式为应用提供资源，那很难实现"按需服务"。本课题从提高云计算数据中心资源利用率为基本出发点，面向复杂应用建立高效的资源配置机制为核心，拟从分析和刻画应用的各功能模式在虚拟化平台上运行时对资源的细粒度需求特征为切入点，构建由基本资源拓扑与不同模式下增量资源需求构成的资源概率需求拓扑。由此拟基于多个应用间增量资源需求的关联关系，研究共享增量资源的应用社区构造及资源需求复合算法，并实现以计算集群为单位的资源规划配置机制。针对集群内的虚拟机部署，基于不同类型应用的性能干扰关系，研究虚拟机均衡资源部署机制。针对运行时应用的演变与整体性能的优化，研究效益最大化的动态资源调整策略，保持系统高效、稳定。上述成果将运行于课题组前期自主研发的云平台原型，验证其正确性。

云计算环境中按需服务的本质目标是要在合适时段为用户虚拟机预留合适的资源，而云端应根据这些需求合理分配资源以保持良好的利用率，从而降低成本。本课题以分析应用系统在数据中心运行时对计算、网络资源的本质需求为切入点，以系统化建立应用感知的动态资源配置机制为目标，设计并实现了面向资源需求动态化应用的虚拟机动态调整机制。提出了网络虚拟化的数据中心动态资源调度技术，实现高效资源部署。针对跨域资源调度问题，提出了面向跨地域应用的多数据中心协同资源调度技术。实现了基于多租户的数据中心网络智能数据流加速技术，并基于该技术设计出基于SDN技术的带宽分配系统FairShare。上述成果运用于我们现有云平台原型，并验证其正确性和有效性。该系统化的资源动态调整机制研究促进了资源利用率的提升，为云服务用户与云平台降低成本。