虚拟计算环境下基于信任的资源调度关键技术研究

发布时间：2018-04-09 15:13

本文选题：虚拟计算环境　切入点：信任　出处：《北京邮电大学》2014年博士论文

【摘要】：虚拟计算环境旨在通过建立在开放的互联网基础设施之上的公用化计算平台,为应用提供和谐、可信、透明的一体化服务支撑。随着虚拟计算技术的发展和应用的深入,一方面,底层资源日趋公用化、规模化、异构化；另一方面,上层应用需求呈现出多样化、差异化、复杂化的趋势。如何有效解决公用环境下海量、异构资源与多样化、多需求目标应用之间的资源调度问题,实现在不可预知的公共运行环境中构造可信服务的目标,日渐成为影响与制约虚拟计算环境发展的关键性问题。本文针对公用环境下的资源调度问题,结合虚拟计算环境实验平台的实际情况,首先在实验平台中引入了信任机制来保证公用环境下资源调度的可信问题,再通过对资源属性和应用需求的逐步深入分析,在现有资源调度技术的研究成果基础上,开展了虚拟计算环境下基于信任的资源调度若干关键技术的研究,以实现资源与应用之间的逐步精细化匹配和调度,从而提升虚拟计算环境的整体性能。本文的创新点主要包括： (1)提出了一种基于贝叶斯信任模型的冗余调度算法——TR Scheduling算法。通过对虚拟计算实验平台上承载的应用做初步分析,应用具有单个任务小、并发量大,任务问相互独立,任务可以以任何顺序执行,任务问不需要通信等特点,并根据虚拟计算环境下海量资源的优势,“以适当的资源换取复杂容错开销”的设计理念实现了基于多副本技术的冗余调度算法。同时,根据贝叶斯信任模型对可调度资源节点的稳定性进行分析计算出资源信任值的排序,给出了按照信任值最大优先(HighFit)算法和信任值最小优先(LowFit)算法两种策略对资源进行动态选择,从而提高了系统的可靠性和资源利用率。 (2)提出了一种实现差异化服务的资源调度模型,也即基于资源预留技术的资源调度模型——RRM模型。通过对应用的需求做初步分析,平台所支撑的应用的从重要程度上可以粗分为“一般应用”和“重要应用”两种类型。RRM资源调度模型依据计算资源属性将普通资源和优质资源分别放入“普通资源池”和“VIP资源池”,并通过动态调整VIP资源比例、动态交换优质与普通资源,合理资源预留策略,该模型在兼顾所有任务调度的基础上,重点提升VIP任务的调度效率与执行成功率,实现了“重要的资源服务于重要的任务”的差异化服务,保障了重要服务等级的应用得到及时响应,有效提升了平台的可用性。 (3)提出了一种实现多条件约束的资源调度模型,也即基于“资源滑动窗口”技术的资源调度模型——RSW模型。通过对应用的需求做进一步分析,平台所支撑的应用需求有着多样化、差异化、复杂化的特点,对所调用资源有多种条件的约束。该模型对资源聚合技术做初步研究,通过对资源静态属性和动态属性的预先分类,使得资源选择的范围逐步缩小,以初步实现“合适的资源服务于合适的任务”的目的,有效提升了平台的适用性。 (4)提出了一种基于模糊聚类的资源调度模型——FCS模型。通过对资源聚合技术做进一步研究,在对资源静态属性分类之后,基于模糊关联聚类算法对资源再做聚类,进一步缩小了资源的选择范围,并降低了直接对原始资源进行聚类的空间复杂度,从而为任务与资源的精确匹配提供了条件,实现了“特定的资源服务于特定的任务”的资源最优配置目的,进一步提升了平台的适用性
[Abstract]:The virtual computing environment through the public based on open Internet infrastructure computing platform for application to provide a harmonious, credible, transparent integration services support. With the development and application of virtual computing technology deeply, on the one hand, the underlying resources become public, scale, isomerization; on the other hand, the upper application showing diverse, differentiated, complex trend. How to effectively solve the public environment massive, heterogeneous resources and diversification, the resource scheduling problem between the needs of the target application, realize the trusted service in the public environment unpredictable in the target, become influence and restrict the development of the key virtual computing environment the problem.
Aiming at the problem of public resource scheduling environment, combined with the actual situation of the virtual computing environment of the experimental platform, in the first experiment platform, introduces the trust mechanism to ensure the credibility problem of public resource scheduling environment, through the attributes of resources and application needs gradually deep analysis, based on the existing research results on Resource Scheduling Technology Study on some key techniques, trust based resource scheduling of virtual computing environment, in order to realize resource and application of the gradual refinement of the matching and scheduling, so as to enhance the overall performance of the virtual computing environment. The main innovations of this thesis include:
(1) proposed a redundant scheduling algorithm based on trust model -- Bias TR Scheduling algorithm. Through the preliminary analysis of the application of bearing virtual computing platform, the application has a single task, with a large amount of tasks are independent of each other, the task can be executed in any order, do not need to ask task communication features. According to the massive resources in virtual computing environment advantage, "design concept to the appropriate resource for complex fault tolerance overhead" realize redundant scheduling algorithm based on multi duplicate technology. At the same time, according to the Bias trust model for scheduling resource node stability analysis to calculate the trust value of resource scheduling, given in accordance with the trust value the maximum priority (HighFit) algorithm and the trust value of the minimum priority (LowFit) algorithm two strategies for dynamic selection of resources, so as to improve the system reliability and resource utilization Rate of use.
(2) we propose a resource scheduling model to realize service differentiation, i.e. resource scheduling model RRM model based on resource reservation technology. Based on the application needs to do a preliminary analysis, application support platform from the important degree can be divided into "general application" and "application" two type.RRM resource scheduling model based on computational resources attribute common resources and high-quality resources were put into the "common resource pool" and "VIP resource pool", and through the dynamic adjustment of VIP resource ratio, dynamic quality and general exchange of resources, reasonable resource reservation strategy, the model based on the consideration of all task scheduling, scheduling and execution efficiency focus on enhancing the VIP mission success rate, to achieve the "differentiated services important resources in the important tasks, to protect the important application service level to respond in a timely manner, effective. The availability of the platform is raised.
(3) proposed a resource scheduling model with multiple constraints, namely "resource based on sliding window technology resource scheduling model -- RSW model. Through the application of the need for further analysis, the application needs to support the platform with diversification, differentiation, characteristics of the complex, a variety of conditions the call of resource constraints. The model do preliminary research on resource aggregation technology, based on the resources of the static and dynamic attributes of the pre classification range makes the resource selection gradually narrowed, with the initial realization of" appropriate resources in the appropriate task ", effectively enhance the applicability of the platform.
(4) we propose a resource scheduling model based on fuzzy clustering FCS model. Further research based on the resource aggregation technology, the static resource attribute classification, fuzzy relational clustering algorithm to do the clustering based on resources, further narrowing the scope of the selection of resources, and reduce the direct clustering of original resources the space complexity of tasks and resources so as to provide accurate matching conditions, to achieve the "objective optimal allocation of resources to specific resources in specific tasks, to further enhance the applicability of the platform

【学位授予单位】：北京邮电大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP393.01

【参考文献】