基于Kubernetes的资源动态调度的研究与实现

发布时间：2018-06-27 20:02

本文选题：云计算 + 容器　；参考：《浙江大学》2017年硕士论文

【摘要】：云计算是一种将计算资源按需供应给用户的新型商业模式,它能满足用户复杂的动态资源需求,从而减少用户在购置基础设施以及硬件维护成本。如何将云平台中的物理资源,如CPU、内存、磁盘和网络带宽等,以最为有效的方式提供给用户,是一个急需解决的难题。目前针对云计算资源的管理和调度研究大都集中在IaaS层,当PaaS平台从IaaS层申请到资源后,如何让资源利用率最大化是至关重要的。对于容器云平台而言,研究重点在于如何去管理和调度运行应用的容器资源。本文立足于容器云平台,对容器层的资源动态调度、实例自动伸缩、负载均衡等方面进行了研究。本文首先研究了具有代表性的开源容器云平台Kubernetes,并详细地介绍了Kubernetes的架构与设计。随后研究了多种可用于预测应用资源使用量的预测模型,重点介绍了 ARIMA模型和神经网络模型,并基于这两种模型提出一种组合模型的方案。根据组合模型的预测结果,提出一种资源动态调度算法,可为Kubernetes集群中Node节点上的应用进行动态的资源调度。并随即设计了基于负载的实例自动伸缩模块、应用对资源敏感程度的负载均衡模块。最后设计了一系列实验,验证了扩展后的Kubernetes的资源动态调度、实例自动伸缩以及负载均衡的能力。实验结果表明,扩展后的Kubernetes能有效提高资源利用率以及应用的服务质量。
[Abstract]:Cloud computing is a new business model which can supply computing resources to users on demand. It can meet the complex dynamic resource requirements of users and reduce the cost of purchasing infrastructure and hardware maintenance. How to supply the physical resources such as CPU, memory, disk and network bandwidth in cloud platform to users in the most effective way is a difficult problem to be solved urgently. At present, most of the researches on the management and scheduling of cloud computing resources are focused on the IaaS layer. When the PaaS platform applies to the resources from the IaaS layer, how to maximize the utilization of resources is very important. For container cloud platform, the research focus is on how to manage and schedule container resources for running applications. Based on the container cloud platform, this paper studies the resource dynamic scheduling, instance automatic scaling, load balancing and so on. This paper first studies the representative open source container cloud platform Kubernets and introduces the architecture and design of Kubernetes in detail. Then we study several forecasting models which can be used to predict the usage of application resources. Arima model and neural network model are introduced. Based on these two models, a combined model scheme is proposed. Based on the prediction results of the combined model, a dynamic resource scheduling algorithm is proposed, which can dynamically schedule resources for applications on Node nodes in Kubernetes cluster. And then the load-based automatic expansion module is designed and the load-balancing module which is sensitive to resource is applied. Finally, a series of experiments are designed to verify the capability of the extended Kubernetes resource dynamic scheduling, instance automatic scaling and load balancing. The experimental results show that the extended Kubernetes can effectively improve resource utilization and application quality of service.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.09

【相似文献】