云环境下多约束动态虚拟资源管理研究

发布时间：2018-01-31 21:15

本文关键词： 云计算资源管理应用可用性资源整合资源安置　出处：《上海交通大学》2013年硕士论文　论文类型：学位论文

【摘要】：本文旨在研究云计算环境中多约束情况下的虚拟资源调度管理问题。多约束一方面是指来自用户的质量属性约束，比如对响应时间和吞吐量等性能要求和对部署在云平台上应用的可用性要求。这些用户约束是以SLA的形式，由用户向云服务提供商提出的。另一方面，多约束是指来自云服务提供商的约束，主要是指减少包括能源开销在内的运营成本的约束。在运行和维护云平台时，云服务提供商总是力求减少能耗来降低成本，提高利润。云平台中资源的供给和部署需要在满足用户约束和降低云服务提供商的运营成本之间进行平衡。为云用户提供更多的虚拟资源可以确保云应用的性能和可用性等需求得到满足，但是会增加云服务提供商的运营成本；反之，如果为云用户提供较少的虚拟资源，尽管可以有效减低云服务提供商的运营成本，但是可能会导致云应用响应时间过长和可用性过低等情况。因此，如何提供、安置和整合云平台上的虚拟资源满足不同利益共享者的约束成为了云服务用户和云服务提供商两方面共同关注的话题。因而，本文提出了一种约束分离的云资源管理框架——通过分离云应用和云服务提供商不同的需求关注点，在不同层次上满足他们各自的需求。我们将云资源的管理分为应用层和平台层上的资源管理。其中，质量属性需求到虚拟资源的映射被抽象成应用控制器。在应用控制器中，用户对性能属性的要求被转化为虚拟资源的数量，将可用性需求转化为虚拟资源位置的部署方案。在平台层的管理框架中，采用整合策略，使得虚拟资源尽可能地运行在较少的物理服务器上，以达到减少运行的物理机数量从而节约能源和成本的目的。在应用层的资源管理模块中，，本文提出了一种计算部署在云平台上的应用可用性的计算模型，以及一种动态虚拟资源安置算法。它们在最大程度上满足了用户的应用可用性需求，同时也减少了分散部署虚拟资源带来的通讯开销。资源安置算法综合使用了两种伸缩策略：水平伸缩和垂直伸缩。垂直伸缩策略是指改变虚拟机内部持有的资源数量，而水平伸缩策略是指改变虚拟机实例的数量。由于这两种策略对应用可用性、通讯开销和软件版本开销的影响不尽相同，我们根据应用当前的可用性和用户约束中规定的可用性的差值来选择执行哪种策略。在平台层的资源管理模块中，本文也给出了一种改进的资源整合方案。该方案将资源的整合分为反应式资源整合模块和周期性资源整合模块。反应式资源整合模块自治地运行在每一台物理服务器内部，根据监测是否有异常情况，决定是否进行资源整合。异常情况包括超过预定资源利用率上限的负载过度情况，和低于预定资源利用率下限的负载不足情况。针对这两种情况本文给出了不同的算法来制定资源整合方案。周期性资源整合模块，作为反应式资源整合的补充，周期性地在可用性区域内进行所有虚拟资源的整合。最后为了验证这两种策略—可用性感知的动态资源安置模块和资源整合模块的正确性和有效性，本文模拟真实云环境的架构，设计并实现了两个实验。在实验一中通过模拟两种典型场景——扩容和减容的应用，对比了水平伸缩策略、垂直伸缩策略和本文提出的可用性感知策略在可用性满足率、平均应用可用性、通讯开销和软件版本开销上的表现。实验二通过模拟一个可用性区域中多台物理服务器和多台多种粒度的虚拟机，来验证改进的资源整合方案的有效性。实验结果表明，本文中提出的算法减少了整合活动带来的迁移成本，是一个实际具有应用价值的有效策略。本文的主要贡献在于从不同利益共享者的角度对多约束资源管理框架的构建进行了初步的探索。将来自不同利益共享者的约束分离开来，在不同层次上予以满足。其中，重点关注了应用层上的可用性约束的满足和云平台层上的节能约束的满足，从而达到云资源使用者和云资源管理者共赢的目的。
[Abstract]:Virtual resource scheduling management problems, this paper aims to study the cloud computing environment with multiple constraints under multiple constraints. One refers to the quality attribute constraints from users, such as response time and throughput performance requirements and deployment of applications on the cloud platform availability requirements. These user constraints are in the form of SLA, proposed by user Xiang Yun service provider. On the other hand, multi constraint refers to from the cloud service provider constraints, mainly refers to reduce energy costs, including operating cost constraints. In the operation and maintenance of the cloud platform, cloud service providers always strive to reduce energy consumption to reduce costs and increase profits. Supply and deployment of resources in the cloud platform the need to balance to meet user constraints and reduce operating costs between cloud service providers. To provide more virtual resources can ensure the performance of cloud applications for cloud users And availability requirements are met, but will increase the cloud service provider operating costs; on the other hand, if the virtual resources provide less for cloud users, although cloud service providers can effectively reduce operating costs, but may lead to a cloud application response time and availability low. Therefore, how to provide, placement and the integration of virtual resources in the cloud to meet the needs of different stakeholders constraints has become a common concern of the two users of cloud services and cloud service providers of the topic.
Therefore, this paper proposes a constraint separation cloud resource management framework, through the separation of cloud applications and cloud service providers to the needs of different concerns, meet their needs at different levels. We will cloud resources into the management of resource management and application layer platform layer. The quality attribute requirements to map virtual resources are abstracted into application controller. In the application of the controller, the user requirements on the performance of property is converted to a virtual number of resources, the availability requirements are mapped into the virtual resource deployment scheme. The position of the platform layer management framework, the integration strategy of the virtual resources as much as possible running on a physical server less, in order to reduce the number of physical machines running so as to save energy and cost.
In resource management module of application layer, this paper presents the calculation model of application availability of a computing deployment in the cloud on the platform, and a dynamic virtual resource placement algorithm. They satisfy the user application usability requirements in the greatest degree, but also reduce the dispersion of the deployment of virtual resources bring communication overhead the integrated use of resources. For algorithm two expansion strategy: horizontal expansion and vertical expansion. The vertical expansion strategy refers to the change of internal resources held by the number of virtual machines, and the horizontal expansion strategy refers to changes in the number of virtual machine instances. By the two strategies of application usability, effects of communication overhead and software version overhead we are not the same, according to the provisions of the current application of usability and user constraints in the availability of the difference to choose what kind of strategy implementation.
In resource management module in the platform layer, an improved resource integration scheme is also given in this paper. The program will be the integration of resources divided into resource integration module and periodic resource integration module. The reactive resource integration module run autonomously on each physical server, according to whether there is abnormal situation monitoring and decide whether the integration of resources. Exceptions include more than the upper limit of the excessive load rate using a predetermined resource, and the shortage of the lower limit of the rate is lower than the load using a predetermined resource. In the light of the two cases presented different algorithms to develop resource integration scheme. The periodic resource integration module, as a complementary of the reactive periodically, the integration of all virtual resources in availability in the region.
Finally in order to verify these two strategies - availability aware dynamic resource placement module and resource integration module is correct and effective, this paper simulates the real cloud environment architecture, design and implementation of the two experiments. Through the simulation of two typical scenarios -- the application of dilatancy and volume reduction in Experiment 1, the level of contrast the expansion strategy, vertical expansion strategy and the availability of knowledge strategy in the availability rate, average application availability, communication overhead and software version overhead on performance. Experiment two through virtual machine simulation of a physical server availability zone and multiple granularities, to verify the effectiveness of resource integration the improved scheme. Experimental results show that the proposed algorithm reduces the cost of migration integration activities, is an effective strategy for the actual application value.
The main contribution of this paper is from different benefit sharing makes a preliminary exploration on the construction of multi constrained resource management framework "perspective. From the separation of the constraints of sharing different interests, to meet the requirements in different levels. Among them, the focus on energy constrained availability constraints on application layer content and cloud platform layer on the meeting, so as to achieve cloud resource users and cloud resource management and win-win goal.

【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP3

【共引文献】