云计算环境下弹性决策机制的研究与实现

发布时间：2018-04-21 14:19

本文选题：云计算 + 弹性　；参考：《上海交通大学》2014年硕士论文

【摘要】：云计算作为一种正在兴起的计算机科学发展方向，在当前社会环境下拥有巨大的潜力。用户通过云平台得到服务和资源，并按其使用付费，使得用户得以以合理的费用满足其需求。云计算之所以能够引起广泛的关注和应用，最直接的驱动力量可以说是经济因素。通过租用云平台中的基础设施与相关软件，用户可以免于为其应用提供基础设施和维护，这大大降低了用户应用的成本。此外，由于云平台还具有动态按需请求分配资源的特点，即弹性提供计算资源的能力，使得用户得以进一步控制其应用代价。这使得云计算成为了更多人的选择。目前许多主流的商业云平台，如Amazon EC2，Windows Azure等，都能够在一定程度上提供弹性支持能力。例如，Amazon EC2的Auto Scaling服务可以与CloudWatch服务相配合以支持Amazon EC2实例数目的动态扩展和收缩。用户可以通过创建触发器(trigger)自定义其应用实例数目动态扩展和收缩的规则。但这种弹性支持能力过于死板和僵硬，无法有效地做出更为精确的判断。这种触发器型的弹性支持能力虽然保证了用户的应用可以得到云平台的弹性支持，但根据实际情况的不同，这种弹性支持能力可能会使用户增加无谓的开销。若用户的应用希望以合理的代价得到较高的性能，一种恰当的资源请求方案是必要的。通常,人工调节的方式往往是能收到良好效果的备选方案。但随着云应用的不断扩大，在不同的负载下，以人工的方式进行判断并为云应用请求资源变得不再可行。为此，构建一个弹性决策机制，使得此机制可以做出决策，保证服务质量并节省开支，是一项有意义且富有挑战性的工作。针对此项问题，本文提出了一项弹性决策机制。此弹性决策机制在支持弹性能力的过程中建立系统负载模型和系统性能模型，并使用这两种模型为用户应用做出弹性决策。与此同时，，为了防止这两种模型失效而导致弹性决策机制丧失其本身的功能特性，此弹性决策机制在动态管理用户应用计算资源的同时，还将定时对系统负载模型和系统性能模型进行有效性检查。根据检查的结果更新或重新建立模型。在文章的末尾，本文将对此弹性决策模型的功能和性能进行实验验证。本文所提出的弹性决策机制将根据实验中给定的系统负载进行弹性功能支持演示。并说明此弹性决策机制在工作中的特点。
[Abstract]:As an emerging computer science development direction , cloud computing has great potential in the current social environment . Users get services and resources through the cloud platform and pay for their use , so that users can meet their needs at reasonable cost .

Cloud computing can lead to a wide range of concerns and applications , and the most direct drive power can be said to be economic . By renting infrastructure and related software in a cloud platform , users can be free from providing infrastructure and maintenance for their applications , which significantly reduces the cost of user applications . In addition , cloud computing becomes a choice for more people because the cloud platform also has the feature of dynamically allocating resources on demand .

Currently , many mainstream business cloud platforms , such as Amazon 2 , Windows Azure , and so on , are able to provide some degree of resiliency support . For example , the Auto Scaling service of Amazon 2 can be matched to a dynamic expansion and contraction of the number of instances of Amazon . By creating triggers , users can customize the dynamic expansion and contraction of the number of instances of their applications . However , this resilient support capability is too inflexible and rigid to make more accurate judgments . This type of resilient support capability , while ensuring that the user ' s application can get the elastic support of the cloud platform , may increase the user ' s meaningless overhead , depending on the actual situation .

An appropriate resource request scheme is necessary if the user ' s application hopes to get higher performance at a reasonable cost . Generally , manual adjustment is often an alternative to good results . However , as cloud applications expand , it is no longer possible to make decisions in an artificial way and to request resources for cloud applications under different loads . To this end , an elastic decision - making mechanism is built so that this mechanism can make decisions , guarantee quality of service and save expenses , a meaningful and challenging task .

In order to prevent the failure of the two models , the elastic decision - making mechanism is used to make elastic decision - making . At the same time , in order to prevent the failure of the two models , the elastic decision - making mechanism loses its own function characteristic . At the same time , the elastic decision - making mechanism will also check the effectiveness of the system load model and the system performance model in order to prevent the failure of the two models .

At the end of this paper , the function and performance of this elastic decision - making model are verified experimentally . The elastic decision - making mechanism proposed in this paper will support the demonstration according to the system load given in the experiment .

【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09

【参考文献】