面向容器化应用的资源管理系统

发布时间：2018-04-23 16:58

本文选题：云计算 + 容器服务　；参考：《华中科技大学》2016年硕士论文

【摘要】：随着以Docker为代表的容器技术兴起,开发者能够将应用封装成标准的容器镜像统一发布到不同的云计算平台。为了部署和编排容器化应用,一些面向容器的资源管理系统相继被推出。这些系统都提供了容器的管理和监控功能,要求用户在提交任务时指定所需的计算资源,然后由系统进行调度。但是在实际的运行过程中,应用工作负载的动态变化使得系统难以及时的调整和满足容器所需的计算资源,可能导致应用性能目标的违背。面向容器化应用的资源管理系统提出了一种综合考虑节点资源利用率和平衡程度的调度算法、一种能够分析和判定容器是否存在性能干扰的检测模型以及一种动态调整容器分配资源的管理策略。系统通过分析容器运行时的资源消耗,动态的调整容器所需的计算资源,能够保证应用的正常运行。具体策略是,系统将应用规划到一个特定的资源池中,而应用包含的容器则共享这个资源池中的资源。系统根据用户在提交任务时指定的资源限制,在调度任务时尽量平衡集群节点的资源利用率,从而避免资源分配不平衡的情况发生。在应用的运行过程中,性能干扰检测模型会监控容器的运行状态,分析计算出容器的性能指标。这些性能指标将作为动态调整容器计算资源的依据,使得受到性能干扰的容器能够回归到正常状态。实验结果表明,面向容器化应用的资源管理系统和当前主流的容器管理系统相比,可以获得良好的性能提升。具体体现在系统的调度策略能够在一定程度上提升集群中计算节点的资源利用率,性能干扰检测模型能够较高概率的检测出容器性能的异常状况。
[Abstract]:With the rise of container technology represented by Docker, developers can distribute the applications into standard container images to different cloud computing platforms. In order to deploy and orchestrate container applications, some container-oriented resource management systems have been introduced. These systems provide container management and monitoring functions requiring users to specify the required computing resources when submitting tasks and then schedule them by the system. However, in the actual running process, the dynamic change of application workload makes it difficult for the system to adjust and meet the computing resources required by the container in time, which may lead to the violation of the application performance goal. A resource management system for container application is proposed, which takes into account the utilization and balance of node resources. A detection model which can analyze and determine whether the container has performance interference or not and a management strategy to dynamically adjust the resource allocation of the container. The system can ensure the normal operation of the application by analyzing the resource consumption of the container and dynamically adjusting the computing resources needed by the container. The specific strategy is that the application is programmed into a specific resource pool, and the container contained by the application shares the resources in the resource pool. According to the resource constraints specified by the user when submitting the task, the system tries to balance the resource utilization of the cluster node when scheduling the task, so as to avoid the imbalance of resource allocation. In the process of application, the performance disturbance detection model will monitor the running state of the container, and analyze and calculate the performance index of the container. These performance indexes will be used as the basis for dynamically adjusting the computing resources of the container, so that the vessel affected by the performance disturbance can return to the normal state. The experimental results show that the resource management system for container application can achieve good performance improvement compared with the current mainstream container management system. The system scheduling strategy can improve the resource utilization of computing nodes in the cluster to a certain extent and the performance interference detection model can detect the abnormal situation of container performance in a higher probability.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP311.52

【参考文献】