面向云计算平台的网络缓存系统的实现

发布时间：2018-08-15 12:49

【摘要】：云计算是一种流行的基于互联网的计算方式。它计算互联网上的硬件资源以及软件资源，并将这两种资源虚拟化成服务交付给个人用户和企业。而云计算系统就是建立在此基础上，它由连接在互联网各种各样的个人计算机、服务器、磁盘整列等设备组成，通过各种虚拟化技术和分布式技术为庞大的用户群体提供各种计算服务和存储服务。在云计算系统中，，存储子系统是通过整合网络上各种不同的存储设备，向用户提供数据存储和管理服务的系统。当今的云存储子系统往往拥有庞大的数据中心，集中管理用户数据，支撑着数量巨大的用户存储服务。这样的基础架构模式带来了宽带瓶颈问题——即当用户访问量过大时，存储子系统由于对外的网络速度问题，无法保证数据传输的速度和质量，影响了存储服务的可用性。本文在阅读国内外相关文献的基础上，设计和实现了基于OpenStack的IaaS基础云计算系统OCloud（OpenStack Based Cloud System），并针对存储子系统的宽带瓶颈问题，设计和实现了基于本地块设备的网络缓存系统NDM-Cache（Network Device Mapper Cache）。本文基于开源的OpenStack项目，设计和实现了基础云计算系统OCloud。首先，本文针对OCloud系统中已有的块级别存储模块存在单点故障的问题，设计和实现了分布式的块级别存储子系统Orthrus。它可以有效地避免单点故障问题。其次，本文为减轻云系统镜像存储模块的空间压力，采用LiveDFS作为存储模块的文件系统，实现了镜像重复数据的删除功能。再次，本文为了能监控系统的所有的虚拟机实例和物理节点，部署了Ganglia监控系统。最后，本文面向校园内师生的科研和学习工作，开发和实现了个人网盘服务，平行计算服务以及教学管理平台服务。全校的师生可以通过web浏览器方便地访问和使用这些服务。针对云计算平台中共享存储子系统的宽带限制问题，本文设计和实现了面向云计算系统的网络缓存系统NDM-Cache。该缓存系统采用本地块设备作为缓存设备。在设计该系统时，本文设计了基于本地块设备的网络缓存系统结构，并提出了面向网络缓存系统的数据管理方案。并且在此基础上，针对内存空间有空余的客户机，本文又提出了双层缓存策略，利用多余内存空间进一步提升系统整体性能。通过实验证明，在8台客户端节点同时访问共享存储设备时，相对于原始共享存储设备和DM-Cache系统，基于本地普通磁盘的原型NDM-Cache系统在数据读取性能上分别平均提高了11.31%和2.34%，而基于SSD盘的原型NDM-Cache则分别平均提高了18.42%和2.76%。并且采用了双层缓存优化策略后，基于普通磁盘的原型NDM-Cache系统比单层系统平均提高了4.57%，而基于SSD盘的原型NDM-Cache系统提高了4.61%。
[Abstract]:Cloud computing is a popular Internet-based computing method. It calculates hardware and software resources on the Internet and virtualizes these two resources into services for individual users and businesses. Cloud computing systems are built on this basis, which are connected by a variety of personal computers, servers, and magnets on the Internet. In cloud computing systems, storage subsystems provide data storage and management services to users by integrating various storage devices on the network. Today's cloud storage subsystems This kind of infrastructure mode brings the broadband bottleneck problem, that is, when the user visits too much, the storage subsystem can not guarantee the speed and quality of data transmission because of the external network speed problem. Availability of storage services.
In this paper, we design and implement the OpenStack Based Cloud System (OCloud) based on IaaS, and design and implement the NDM-Cache (Network Device Mapper Cache) based on local block devices to solve the bottleneck problem of the storage subsystem.
Based on the open source OpenStack project, this paper designs and implements the basic cloud computing system OCloud. Firstly, this paper designs and implements the distributed block-level storage subsystem Orthrus, aiming at the problem of single point of failure in the existing block-level storage module in the OCloud system. Thirdly, Ganglia monitoring system is deployed to monitor and control all virtual machine instances and physical nodes of the system. Finally, this paper focuses on the research and learning of teachers and students on campus. As a result, we have developed and implemented personal network disk services, parallel computing services and teaching management platform services. Teachers and students can access and use these services conveniently through web browsers.
In view of the bandwidth limitation of shared storage subsystem in cloud computing platform, this paper designs and implements a network caching system NDM-Cache for cloud computing system. This caching system uses local block devices as caching devices. On this basis, a double-layer cache strategy is proposed for clients with spare memory space to further improve the overall performance of the system. Experiments show that compared with the original shared memory, when eight client nodes access the shared storage device at the same time For storage devices and DM-Cache systems, the NDM-Cache system based on local ordinary disk improves the performance of data reading by 11.31% and 2.34% respectively, while the NDM-Cache system based on SSD disk improves by 18.42% and 2.76% respectively. The layer system increased by 4.57% on average, while the prototype NDM-Cache system based on SSD disk increased 4.61%..
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333

【参考文献】