面向SLA的服务组合QoS管理及优化技术研究

发布时间：2018-05-28 21:28

本文选题：服务等级合约 + 服务组合　；参考：《复旦大学》2014年博士论文

【摘要】：本文关注服务组合场景下面向SLA(service level agreement)的QoS管理与优化问题。在开放的服务计算环境下,服务组合方需要有效的QoS管理机制,通过对其运行环境的灵活配置与优化满足前端多用户的动态QoS需求。另一方面,服务双方跨管理域的业务关系由SLA及其管理机制来规范和保证,作为一种声明式合约,SLA反映了服务使用方的QoS需求并为服务提供方的QoS管理提供了依据和基础,因此需要在SLA框架下考虑和设计服务组合方QoS管理的功能结构和管理流程,其中服务组合前端QoS需求由前端SLA合约的签订情况决定,而后端SLA合约则保证了各伙伴服务的服务质量。由于服务组合其前后端QoS属性之间的关联性,其前端所能支持和保证的服务质量不仅与底层计算资源配置有关,而且取决于后端各伙伴服务的SLA合约签订等级。因此面对前端多用户的动态SLA签订情况,服务组合方需要实现前后两端SLA合约的动态协调并同时完成内部计算资源的优化配置,才能在满足前端QoS需求的同时最大化其SLA收益,而这两方面资源的优化配置依赖于有效的优化方法。本文充分考虑现有的SLA管理框架及技术,在前后端一致的SLA环境下,本文提出一个面向SLA的服务组合QoS管理模型。模型中首先针对服务组合及其运行环境中两方面的可配置资源,形式化定义了服务组合QoS管理中两个核心的多目标优化问题,并进一步建立了二者之间的关系,使其可以在统一的优化方法下解决。随后本文提出一个面向SLA的服务组合QoS管理模型,在SLA合约管理层、服务组合实例管理层以及计算资源配置层三个管理层次分别给出了其管理实体、功能构成以及主要的管理活动,并在配置方案优化的基础上建立了各层次功能以及管理实体之间的对应关系,构成完整的系统模型。最后在分级和定制两种管理方式下,给出了整个系统的动态配置流程。为了高效得到满足前端QoS需求的最优配置方案,本文针对服务组合QoS管理中的这一核心问题,分别在考虑偏好和不考虑偏好两种情况下给出了两种多目标优化方法及算法。当不考虑各个QoS维度上的偏好时,针对现有单纯基于帕累托评价的优化方法中结果集规模不可控问题,引入多目标决策辅助方法：PROMETHEE,从而在不提供偏好的情况下实现对候选配置方案的高区分度量化评价,在此基础上上提出了遗传算法：P_MOEA,高效返回top-k个PROMETHEE意义上的帕累托最优配置方案。在考虑偏好时,针对现有研究中广泛使用的线性效用函数方法存在的不足与局限,引入Tchebycheff距离作为服务组合QoS优化问题的标量化函数,从而根据偏好设置定位其最优配置方案；随后针对偏好设置的困难,给出一个基于模糊谓词的偏好模型,并提出偏好设置过程将模糊谓词表示的偏好信息转化为标量化函数定义需的量化偏好权重；最后提出了两个具有不同优化目标和特点的遗传算法：single_EA与hybrid_EA,前者返回单个最优配置方案并且效率较高,适合于运行时的配置方案优化过程。后者返回top-k个最优配置方案集合,支持交互式决策过程,适合于配置方案的预先规划与创建。本文通过详细的实验研究验证了算法的性能。
[Abstract]:This article focuses on the QoS management and optimization of the SLA (service level agreement) under the service composition scenario. In an open service computing environment, the service composition party needs an effective QoS management mechanism to meet the dynamic QoS requirements of the front-end multiuser through the flexible configuration and optimization of its running environment. On the other hand, the service is cross managed by both parties. The business relationship of the domain is regulated and guaranteed by SLA and its management mechanism. As a declarative contract, SLA reflects the QoS requirements of the service users and provides the basis and foundation for the QoS management of the service provider. Therefore, the functional structure and management process of the QoS management of the service composition party should be considered and designed under the SLA framework, in which the service group is in the service group. The QoS requirements for the front end are determined by the signing of the front end SLA contract, while the back end SLA contract guarantees the service quality of the partners. The service quality that the front end can support and guarantee is not only related to the underlying computing resource configuration, but also depends on the partner service at the back end because of the association between the front and back end QoS attributes of the service composition. In the face of the dynamic SLA signing of the front end multiuser, the service composition party needs to realize the dynamic coordination of the SLA contracts at both ends and complete the optimization configuration of the internal computing resources at the same time, so as to maximize the SLA revenue at the same time to meet the needs of the front-end QoS, and the two aspects of the optimization of the resources are dependent on the two aspects. This paper takes full consideration of the existing SLA management framework and technology, and proposes a SLA oriented QoS management model for the service composition in the same SLA environment with the same front and back end. The model firstly defines two configurable resources in the service composition and its running environment, and defines the two cores of the QoS management of the service composition. The problem of multi-objective optimization and the relationship between the two are further established so that it can be solved under the unified optimization method. Then a SLA oriented service combination QoS management model is proposed, and the management reality is given in the SLA contract management layer, the service composition instance management layer and the three management levels of the computing resource configuration layer respectively. Body, functional composition and main management activities, and on the basis of the optimization of the configuration scheme, set up the functions of all levels and the corresponding relationship between the management entities, and form a complete system model. Finally, the dynamic configuration process of the whole system is given under two management modes of classification and customization. In order to efficiently meet the needs of the front end QoS, In this paper, two multi-objective optimization methods and algorithms are given under two conditions of preference and without preference. When the preference on each QoS dimension is not considered, the size of the result set in the existing Dan Chunji's Pareto evaluation method is not considered. In the uncontrollable problem, the multiobjective decision assistant method is introduced, PROMETHEE, to realize the quantized evaluation of the high area diversity of the candidate allocation scheme without providing preference. On this basis, the genetic algorithm is proposed: P_MOEA, which returns the Pareto optimal allocation scheme in the sense of Top-k PROMETHEE efficiently. There are shortcomings and limitations of the linear utility function method widely used in the study. The Tchebycheff distance is introduced as the scalar function of the service combination QoS optimization problem, and the optimal allocation scheme is located according to the preference setting. Then, a preference model based on the fuzzy predicate is given for the difficulty of preference setting, and the bias is put forward. The good setting process converts the preference information expressed by the fuzzy predicate to the quantized preference weight required by the scalar function definition. Finally, two genetic algorithms with different optimization objectives and characteristics are proposed: single_EA and hybrid_EA, the former returns to a single optimal configuration scheme and has a higher efficiency, and is suitable for the optimization of the configuration scheme at run time. The latter returns to the Top-k optimal configuration set, supports the interactive decision-making process, and is suitable for the pre planning and creation of the configuration scheme. This paper validates the performance of the algorithm through a detailed experimental study.
【学位授予单位】：复旦大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP393.09

【相似文献】