面向云服务的弹性调度算法的研究与实现

发布时间：2018-03-27 05:07

本文选题：云服务　切入点：弹性　出处：《哈尔滨工业大学》2017年硕士论文

【摘要】：云服务是一种基于互联网的服务模式,通常包括了从服务申请、使用到支付的全部过程。云计算的吸引力就在于其计费标准是按照用户实际使用的资源付费的,云服务有能力在任何时间灵活的增加或者减少资源以满足用户的需求。弹性作为云计算模式下的一个重要特征,资源的动态分配机制实现在需求增加时,相应的增加资源;在需求降低时,将分配的资源回收。弹性通过按需付费的模式灵活适应用户波动的需求,在保证服务等级协议(SLA,Service Level Agreement)和服务质量(Qo S,Quality of Service)的前提下使资源供应与资源需求尽可能接近。弹性调度是实现云计算系统中动态、频繁以及自动变更资源配置关键,其性能很大程度上决定了云计算系统提供服务的能力。传统面向可扩展性的调度因无法随着系统资源需求的变化而动态伸缩资源,导致大量的资源浪费,以致无法满足当前云服务的要求。一个可以快速伸缩、精确配置的弹性调度策略将会有效的提升云计算系统的性能。但是目前的弹性算法主要还存在两方面的问题:一方面体现在面临大规模突发式与激增式应用负载涌现时,出现调度时延长(传统方式下虚拟机从镜像启动到服务可用的时间通常在5~15分钟左右)、大量Qo S不满足以及导致SLA违约等问题;另一方面是目前的弹性调度算法大多只从用户或者服务提供者单方面进行优化,很少有从双方共同的角度,以费用最低为原则,以寻求Provision-Cost平衡为目标进行研究。为解决上述问题,本文通过对调度算法进行深入研究,提出了一种结合反馈与多级预测机制的敏捷弹性在线调度算法,采用预测机制提前预留资源有效避免了由于资源延迟而导致的服务违约问题;加入预测式补偿机制,在资源分配过程中结合系统反馈信息,在保证预测结果准确性的同时提高资源分配的合理性。文章从资源配置速度与配置精确度两个方面出发,既考虑到用户的经济效益又结合服务提供者的切身利益,在保证Qo S和SLA的基础上,以最小的成本实现最大的资源利用率,并设计罚金模型作为评测指标进行验证。最后,在Cloud Stack云平台上,部署适用于云环境下的弹性应用,提供多种资源、多种负载验证本文算法的有效性。
[Abstract]:Cloud service is an Internet-based service model that typically includes the entire process from service application, usage to payment. Cloud computing is attractive because its billing standards are based on the resources actually used by the user. Cloud services have the ability to flexibly increase or reduce resources at any time to meet the needs of users. Flexibility as an important feature of cloud computing mode, the dynamic resource allocation mechanism is implemented when the demand increases, the corresponding increase in resources; When demand decreases, recycle allocated resources. Flexibility adapts to fluctuating user demand through a pay-on-demand model. On the premise of guaranteeing the service level agreement (SOA) and quality of service (QoS), the resource supply and resource demand are as close as possible. Flexible scheduling is the key to realize dynamic, frequent and automatic resource configuration in cloud computing system. Its performance largely determines the ability of cloud computing systems to provide services. Traditional scalable scheduling can not scale resources dynamically with the change of system resource requirements, resulting in a large amount of waste of resources. Unable to meet the requirements of the current cloud service. The flexible scheduling strategy with precise configuration will effectively improve the performance of cloud computing systems. However, there are still two main problems in the current elastic algorithms: on the one hand, it is reflected in the emergence of large-scale burst and surge application load. The scheduling time is prolonged (in the traditional mode, the time of virtual machine starting from mirror to service is usually about 515 minutes, a large number of QoS is not satisfied and leads to SLA default and so on; On the other hand, most of the current flexible scheduling algorithms are optimized unilaterally by users or service providers, and few of them are based on the principle of minimum cost from the common point of view. In order to solve the above problems, this paper puts forward an agile elastic online scheduling algorithm combining feedback and multilevel prediction mechanism. The use of predictive mechanism to reserve resources in advance effectively avoids the problem of service default caused by the delay of resources, and adds a predictive compensation mechanism to integrate the feedback information of the system in the process of resource allocation. In order to ensure the accuracy of prediction results and improve the rationality of resource allocation, this paper considers the economic benefits of users as well as the vital interests of service providers from the two aspects of resource allocation speed and allocation accuracy. Based on the guarantee of QoS and SLA, the maximum resource utilization is realized at the minimum cost, and the fine model is designed as the evaluation index. Finally, on the Cloud Stack cloud platform, the flexible application is deployed to the cloud environment. Provide a variety of resources and loads to verify the effectiveness of the algorithm.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.09

【参考文献】