云系统节点与网络资源管理机制的研究与实现

发布时间：2018-04-09 01:02

本文选题：资源管理　切入点：协同监控　出处：《南京邮电大学》2017年硕士论文

【摘要】：随着云计算不断发展,数据中心承载的交换机和服务器等硬件设备越来越多,海量的资源带来了巨大的管理压力,威胁着云平台的健康、稳定运行。在云资源管理中一直存在着两个问题,一方面为了协调任务的部署,提高系统的执行效率和稳定性,云系统需要高效的节点资源监控机制。然而传统的集中式监控架构容易导致单点失效和性能瓶颈等问题,不能适应大规模的云环境。另一方面,复杂且繁重的业务为云数据中心带来了沉重的网络负载,同时网络负载不均衡导致了资源利用率低、时延长、灵活性与可靠性差、吞吐率不理想等问题,传统的网络架构或简单增加新的网络资源等方式不能有效解决问题,因此为有效提高云数据中心网络资源利用率和服务质量,降低运营成本,特别是解决网络部分拥塞、部分利用率低以及可靠性等问题,亟需构建一套适应于云系统的网络管理机制。针对这两个问题,本文从节点资源监控角度提出了一种新型的分布式协同监控机制,包括一种分布式协同监控模型和一种自适应监控阈值控制算法,从网络资源管理角度提出了基于SDN(Software Defined Network)的多路径流量调度机制。具体的研究工作如下:(1)提出了一种面向云计算系统的分布式协同监控模型(Distributed Collaborative Monitoring Model,DCMM),以数据节点相互感知、彼此监控的方式,实现正常状态下数据节点的自我管理,异常状态下向主节点的及时信息推送,从而均衡监控负载,避免单点失效和性能瓶颈。(2)提出了一种自适应阈值控制算法(Adaptive Threshold Control Algorithm,ATCA),基于历史监控数据,动态地调整阈值以识别没必要推送至监控节点的重复监控数据,进而减少监控数据传输,减轻监控系统对整个系统的影响。(3)提出了一种面向云系统网络的基于SDN架构的多路径流量调度机制(Multi-path Traffic Scheduling mechanism based on SDN,MTSS),利用网络负载均衡算法为新加入的数据流选择负载最轻的路径,并周期性监控网络各链路带宽利用情况,自适应地根据链路负载情况,充分利用网络中空闲链路,制定合适的路由转发路径,调度负载较重链路上的流量,实现灵活的可编程式数据转发以均衡网络负载。MTSS能够有效提高了网络的负载均衡性、资源利用率,从而有效降低网络丢包率、时延,提高网络的可靠性和吞吐率。(4)在协同监控和流量调度机制的基础上,设计并构建了云资源管理系统,详细阐述了系统的架构,并分别根据系统界面介绍了各功能和具体效果。
[Abstract]:With the continuous development of cloud computing, more and more hardware devices such as switches and servers are loaded in the data center. Massive resources bring huge management pressure, threatening the health of cloud platform and running stably.There are two problems in cloud resource management. On the one hand, in order to coordinate the deployment of tasks and improve the efficiency and stability of the system, the cloud system needs an efficient monitoring mechanism of node resources.However, the traditional centralized monitoring architecture can easily lead to single point failure and performance bottlenecks, and can not adapt to large-scale cloud environment.On the other hand, the complex and heavy business brings heavy network load to the cloud data center. At the same time, the network load imbalance results in low resource utilization, prolonged time, poor flexibility and reliability, poor throughput and so on.Traditional network architecture or simply adding new network resources can not effectively solve the problem. Therefore, in order to effectively improve the utilization of network resources and the quality of service of cloud data center, reduce the operation cost, especially solve the network congestion.Due to some problems such as low utilization and reliability, it is urgent to construct a network management mechanism suitable for cloud systems.Aiming at these two problems, this paper proposes a new distributed cooperative monitoring mechanism from the point of view of node resource monitoring, including a distributed cooperative monitoring model and an adaptive threshold control algorithm.A multi-channel runoff scheduling mechanism based on SDN(Software Defined Network is proposed from the point of view of network resource management.The specific research work is as follows: (1) A distributed Collaborative Monitoring Model DCMMN (distributed Collaborative Monitoring Model) for cloud computing system is proposed to realize the self-management of data nodes under normal condition by mutual perception and mutual monitoring of data nodes.In order to balance the monitoring load and avoid single point failure and performance bottleneck, this paper presents an adaptive Threshold Control algorithm based on historical monitoring data.Dynamically adjust the threshold to identify repetitive monitoring data that is not necessarily pushed to the monitoring node, thereby reducing the transmission of monitoring data,In this paper, we propose a multi-path Traffic Scheduling mechanism based on SDN MTSS scheduling mechanism based on SDN architecture for cloud system network, and use the network load balancing algorithm to select the new data stream. (3) to reduce the impact of monitoring system on the whole system, we propose a multi-path Traffic Scheduling mechanism based on SDN MTSS scheduling mechanism based on SDN architecture for cloud system network.The lightest load path,And periodically monitor the use of each link bandwidth, adaptively according to the link load, make full use of the free link in the network, make appropriate routing and forwarding path, and schedule the traffic on the heavy load link.The realization of flexible programmable data forwarding to balance network load. MTSS can effectively improve the load balance and resource utilization of the network, thus effectively reduce the packet loss rate and delay.On the basis of cooperative monitoring and traffic scheduling mechanism, the cloud resource management system is designed and constructed. The architecture of the system is described in detail, and each function and concrete effect are introduced according to the system interface.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.07

【参考文献】