【作者】 徐磊；

【导师】 吴澄；

【作者基本信息】 清华大学 ， 控制科学与工程， 2014， 硕士

【摘要】 云计算技术在工业界的成熟为传统服务科学带来了新的思路。狭义的“云”通常等同于“数据计算云”，即通过互联网向大众提供计算资源共享的系统，具有按需提供服务、规模化、可伸缩等特点。本文基于“数据计算云”的概念与特点，将其扩展至“抽象服务云”的使用场景中，为更一般化的服务系统架构提供了描述。云环境中，对服务资源的管理，特别是服务预约环节的优化调度直接影响整个系统的服务效率和效益。传统调度方法仅限于用户与服务提供商间的一对一交互，缺少用户作业间、服务资源间的相互联系；未考虑资源本身的复杂性，将服务能力简单等同于单资源的可用时段。本论文在云环境中对服务系统进行了重构，采用服务能力抽象化、资源虚拟化、服务注册中心等技术，并采用云计算中普适性强的方法对系统进行了优化，提高了系统的综合服务效率和效益指标。论文主要工作包括：1.综述了云环境的定义与特点，调度问题涉及的基本概念，以及资源调度与作业调度的关系。针对传统预约服务调度的不足，设计方案将其扩展至云环境中，并总结出可推广的策略，对服务资源的抽象过程和作业的安排过程予以精确描述。2.面向资源调度，研究多维度资源的简化方法，建立对应数学模型，并在此基础上设计基于多维聚合的虚拟资源调度算法。通过仿真实验，在不同特征的参数下，客观评估各类算法的性能优劣，给出调度方案选择的建议。3.面向作业调度，研究跨云平台的作业调度策略，设计作业时间约束下的注册中心服务能力调度模型。将作业花销定义为服务价格和地理迁移价格的组合，通过全局最优化保证整个系统的效益。4.文中的调度模型在“211”工程项目CERS的仪器平台校际互通模块中得到利用。基于以上算法和模型，实现了分析测试预约服务的原型系统，为新一期项目的功能开发提供了参照。同时也可为类似的服务系统提供一定的借鉴作用。更多还原

【Abstract】 The cloud computing technology developed in industry is introduced to service science and engineering as a new approach. The traditional cloud, which always refers to the data computing cloud, provides shared computing resources to public via internet according to their demand. This kind of cloud is characterized by its large-scale supply, highly scalable and on-demand services, which is similar to other common service systems.In order to represent a general framework for different service system, this study proposes a general service cloud framework as an extension of data computing cloud.In the cloud environment, the service resource management, especially the reservation scheduling is of great importance to the service efficiency and effectivity of service system. The traditional reservation system just supports one-to-one interaction between user and service provider, regardless of the association between users’jobs, or service resources, moreover, it defines the service ability just according to the time availability of single resource and disregards the complexity of resources. In order to deal with these problems, the system may be reconfigured and deployed backward in cloud environment, in addition, by means of service ability abstraction, resources virtualization and service registry, the efficiency and effectivity of the comprehensive service system should be improved and optimized accordingly, and thus, some widely used cloud computing methods can also be employed in system optimization. This thesis would be organized as following sections.Chapter2introduces the definition and features of cloud environment, scheduling related concepts, as well as the relationship between resource allocation and job scheduling. In concern of the insufficient reservation problem in traditional service environment, this study proposes a transferable scheme to represent the abstraction process of service resources and allocation of jobs in cloud environment.With respect to resources allocation, Chapter3develops a model to simplify the abstraction process for multidimensional resources. Several allocation algorithms for the multidimensional resources have been developed accordingly, and further compared using a simulation case with different model parameters. Consequently, alternative allocation strategies are recommended under different conditions.With respect to job scheduling, Chapter4focuses on the scheduling strategy across cloud platforms. A service ability scheduling model is proposed for registration center with constraints of time consumption. In the model, the cost of jobs is calculated by a combination of service price and transportation cost, which can thus be optimized globally to achieve better system benefit.Finally, Chapter5demonstrates the applications of proposed allocation and scheduling models in the "211" engineering project of China Equipment and Education Resource System (CERS). The prototype system for the scheduling, analysis and assessment of equipment reservation module is realized finally based on the proposed models and related algorithms, which can be further applied in the future project and other related service systems.更多还原