【作者】 杨成伟；

【导师】 孟祥旭；

【作者基本信息】 山东大学 ， 计算机软件与理论， 2012， 博士

【摘要】 近两年,大规模实例密集型工作流应用需求不断增加,为工作流技术提供了更加广阔的需求空间。在制造领域中的一些工作流不仅规模大、实例密集,而且实例之间还存在一定的依赖性。这使当前工作流系统从应用范围和能力上,都难以满足现代企业的需要。云工作流的提出为解决当前应用中大规模密集型应用提供了技术手段。一方面,大量的云服务能够满足工作流执行的要求,最大程度满足用户的需要；另一方面,大量的云服务能够通过工作流组织起来,用户可以通过可视化建模方式组织自己所需的业务流程。云工作流服务本身和其它服务一样,都是通过服务定制的方式给用户提供服务的。在工作流初始化阶段,用户首先在云计算平台上签署云服务使用契约,提交业务流程的定义,搜索匹配所需云服务,并与流程的定义进行绑定。在工作流执行阶段,工作流引擎将根据用户指定的QoS限制,采用优化策略进行负载均衡调度。如果在此过程中出现QoS冲突,工作流引擎会根据预先的设定对冲突进行处理。在这一过程中,工作流的运行环境发生了巨大变化。用户对云工作流系统的性能、效率、安全等非功能性方面要求很高。而云工作流系统必须能够支持和满足这种需要,在其生命周期内满足不同组织和用户的个性化QoS定制要求。为适应这种改变,工作流中的一些典型问题,需要被重新提出,如体系结构问题,调度问题,资源管理问题等。云工作流系统构建关键技术的研究得到了国家863高新计划项目《支持装备制造产业集群业务协同服务支持平台》和《业务关联的中小企业群信息化服务平台开发与应用》等科研项目的资助,以山东省制造业信息化服务平台为依托,在平台原有功能基础上,通过扩展流程服务组件,使其能够满足云计算环境下用户对工作流服务的个性化QoS定制需要。结合制造领域的应用背景,对云工作流系统体系结构、生命周期、组合流程、任务调度、资源管理等方面进行了初步探讨。本文所做工作的主要贡献包括四个方面：1.提出了面向实例密集型应用的云工作流系统体系结构云工作流系统利用云计算提供的基础设施服务,组织、协同云服务提供商所提供的各种类型的服务,满足云计算环境下实例密集型应用需求。因此,将云工作流系统与云计算平台体系结构各层进行对应,提出了云工作流系统体系结构。同时,将云工作流的生命周期分为四个过程,其中建模与仿真过程在工作流构建阶段进行,搜索匹配及调度执行在工作流运行阶段进行,交易与评价过程在工作流完成阶段进行的。2.提出了基于动态流程的服务组合协同模型云工作流系统引擎在工作流执行阶段将用户的流程服务请求优化调度到云服务提供商提供的服务资源上执行。由于云计算环境下的服务提供商所发布的服务具有异构性,需要在平台上提供统一的封装、资源选择与绑定标准。因此,提出了支持服务协同的动态组合流程方法,包括业务功能建模、服务搜索匹配、服务动态绑定,并针对组合流程中服务依赖关系,提出了对服务依赖度的验证算法。3.提出了基于QoS优化的任务预调度模型与算法由于采用了按需付费的服务模式,云工作流需要为用户提供高服务质量(QoS)。基于QoS感知的工作流调度方法,加入了对服务QoS条件的限制。在服务层调度中,采用改进的全局遗传演化方法,对工作流任务进行预调度,提高了工作流执行阶段的系统吞吐率。4.提出基于负载感知的资源管理与延迟调度策略与算法为了确保在工作流执行阶段服务资源充足,提升平台的峰值负载能力,满足每个实例任务的QoS限制。在任务层的调度中,提出了基于负载感知的延迟调度策略,重点研究了调度时机选择问题,给出了延迟调度算法,对比了调度过程中的资源的消耗情况。基于上述研究工作,在山东省制造业服务平台(SDMSP)基础上设计和研发了云工作流系统原型(I2-CWS),并对其应用案例进行了讨论。本文仅解决了云工作流系统上的部分问题,未来将在进一步完善平台基础上,对云工作流变更、互操作、冲突处理等问题展开研究。更多还原

【Abstract】 Nearly two years, the demand of large scale instance intensive workflow application is increasing, which provides workflow technology for more vast demand space. In the manufacturing, some workflow instances are not only large scale and instance intensive, and still, there is dependence between them. This makes the current workflow system cannot meet the needs of modem enterprise. Cloud workflow provides the technology means to solve the problem of large scale workflow application. On the one hand, a lot of cloud services can meet the requirement of workflow execution, and satisfy the needs of users. On the other hand, it makes the organization of many cloud services through workflow. Users can custom their business process through the virtual modeling tools.Workflow service in the cloud computing environment is the same as other services. It provides service through service customized way too. All services must follow the market oriented mode, and pay for use. Workflow service is no exception. At workflow initial stage, users firstly sign the contract in cloud computing platform, submit the definition of workflow, search and match cloud services, and then bind the process definition. At workflow execution stage, workflow engine will adopt the optimized scheduling strategy according to user’s QoS. If QoS conflict is appeared, workflow engine will process the conflict according to the configuration in advance.In this processing, the workflow running environment has been large changed. The requirements of user’s non-function, such as performance, efficiency, as well as safety are higher. And the requirements of user’s personalized QoS service customization must be satisfied and supported by the cloud workflow platform in the lifecycle. Therefore, in order to adapt this change, some typical problems of workflow must be put forward, such as architecture, scheduling, resource management, etc.The research of key issues under cloud workflow platform is supported by the national863high and new project "The collaborative service platform funded for equipment manufacturing industry cluster" and "The information service platform for business associated small and medium sized enterprises", etc. Based on the functions of Shandong manufacturing industry information service platform, it extends more functions and satisfies the needs of personalized QoS customization. In the background of manufacturing industry, some key issues such as cloud workflow architecture, lifecycle, composition process, task scheduling and resource management are discussed.The main contributions of the thesis are as follows:1. Propose the cloud workflow system architcture of instance intensive application orientedCloud workflow system in the workflow system platform not only uses the infrastructure services provided by cloud computing environment, but also collaborate the other services which is provided by other service providers. Therefore, this paper maps the cloud workflow system architecture into cloud computing platform, and put forward cloud workflow system architecture. Then, we put forward the four processes of cloud workflow lifecycle, including modeling and simulation process in workflow initial stage, searching, matching, scheduling and executing process in workflow execution stage, and trading and evaluation process in workflow complete stage.2. Propose the service collaborative model based on dynamic composition process.Cloud workflow engine will be responsible for quickly searching the services and scheduling those services to the service resources that provided by cloud service providers. Because cloud computing environment service provider’s services are heterogeneous, it needs to provide a uniform encapsulation, resource selection and binding standards. Therefore, we put forward the method of dynamic service composition process, including business function modeling, service searching and matching, service dynamic binding. According to the relationship of service dependence, we put forward the dependency validation algorithm.3. Propose the QoS based global optimized task scheduling model and algorithm.Due to adopting the pay it on demand mode, users in cloud computing environment have high needs of QoS. The QoS aware workflow scheduling method includes the conditions of the QoS constraints. In the service layer, it uses improved genetic algorithm as pre-scheduling method to improve the throughput in workflow execution stage. 4. Propose load-aware based resource management and delayed scheduling strategy and algorithmIn order to ensure the enough resource in workflow execution stage and improve the peak load capability of the platform. In the task layer, it puts forward load-aware based delayed scheduling strategy and focuses on the problems of scheduling chance and present the delayed scheduling algorithm.On the above research works, we design and develop cloud workflow system prototype (I2-CWS) based on Shandong manufacturing industry information service platform (SDMSP), and discuss the application cases. In this paper, we only solve some parts of problems in cloud workflow platform. Other problems such as workflow change and interoperability will be as the further research works.更多还原