H3Cloud高可靠性和高可用性技术白皮书

H3Cloud高可靠性和高可用性技术白皮书Copyright©2012杭州华三通信技术有限公司版权所有，保留一切权利。非经本公司书面许可，任何单位和个人不得擅自摘抄、复制本文档内容的部分或全部，并不得以任何形式传播。本文档中的信息可能变动，恕不另行通知。i目录1技术应用背景···················································································································12H3Cloud实现的技术特色···································································································22.1H3Cloud云计算软件简介································································································42.2相关技术基础简介··········································································································22.2.1共享存储·············································································································22.2.2H3CloudvMigration·······························································································32.3H3Cloud高可靠性（HA）技术·························································································52.3.1相关术语·············································································································52.3.2物理服务器主机HA工作原理··················································································62.3.3虚拟机HA工作原理······························································································62.3.4技术特色总结·······································································································72.4H3Cloud高可用性技术···································································································82.4.1动态资源调整·······································································································82.4.2虚拟机资源限额··································································································102.5应用限制····················································································································123典型组网案例·················································································································123.1组网拓扑····················································································································123.2注意事项····················································································································133.2.1对服务器硬件的要求····························································································133.2.2整合比（单台服务器上虚拟机数量）的决定因素························································144参考文献·······················································································································1411技术应用背景随着虚拟化和云计算浪潮在全球IT行业的兴起，越来越多的企业、行业和运营商纷纷将自身的IT架构切换到虚拟化环境中。虚拟化技术对数据中心内未被充分利用的服务器进行整合，极大地降低了客户的一次性投入成本，精简了数据中心物理服务器的数量，同时，减少了供电、制冷、场地和运维人员方面的运营成本。但是，虚拟化也为IT应用带来了单点故障问题，在未实施虚拟化技术之前，IT管理员往往遵循“根据最坏情况下的工作负载来确定所有服务器的配置”这一策略，即一台高性能物理服务器仅安装一个应用程序。在这种情况下，即使该物理服务器出现了断电或操作系统崩溃等异常状况，最多只会影响到一个应用的运行，而在虚拟化环境下，每台物理服务器往往运行多个虚拟的应用服务器，因此，虚拟化技术的实施将使IT环境面临的灾难破坏性更严重，尤其对于一些重要的业务入口或接入点（如企业的生产服务器和金融行业的数据库服务器等），即使出现秒级的业务中断，也将遭受灾难性的后果。在这种应用背景下，如何保证虚拟化环境下业务应用的高可靠性和高可用性，成为急需解决的一个技术问题。VMVMVMVMVMVMVMVMVMIPnetwork图1物理服务器故障造成虚拟化业务全部中断传统的集群解决方案（如微软的ClusterService和VeritasClusterServer）致力于在发生服务器主机故障或虚拟机故障时，在最短的应用程序停机时间内实现即时恢复，要达到这个目标，IT基础架构必须进行如下设置：每台物理服务器和虚拟机都必须有一个镜像虚拟机（可能在其它服务器主机上）。使用集群软件将服务器（或虚拟机及其主机）设置为互相镜像，一般情况下，由主虚拟机向镜像发送心跳信号，一旦发生故障，镜像将立即接管。下图显示使用传统集群方法的典型的虚拟机设置：2VMVMVMVM机箱内集群跨机箱集群图2使用传统集群配置确保虚拟化应用的高可靠性设置和维护这样的集群解决方案成本昂贵，并会占用大量资源。每添加一台新的虚拟机，可能需要更多的虚拟机甚至是更多的服务器主机才能进行故障切换。您将不得不设置、连接和配置所有的新计算机，并更新群集应用程序的配置。总之，传统解决方案虽然能够保证快速恢复，但却是资源和劳动密集型的，而且通常存在对应用程序和操作系统的依赖性问题。由于集群解决方案的成本和复杂性问题，它们通常只被一小部分企业应用程序所采用，而大部分应用程序则没有任何故障切换保护。为了解决虚拟化和云计算部署带来的单点故障问题，H3Cloud云计算软件摒弃了高成本、高复杂度的传统集群方案，通过为应用程序提供可用的、经济的高可用性，而使其更“大众化”。2H3Cloud实现的技术特色2.1相关技术基础简介2.1.1共享存储H3Cloud云计算管理平台中的存储用于保存虚拟机的操作系统、应用程序文件、配置文件以及与活动相关的其它数据，是虚拟机正常工作的基本前提条件。根据存储的种类不同，可以分为本地存储和共享存储两种。在部署了H3Cloud云计算管理平台，并将主机作为被管理资源对象添加到H3Cloud云计算管理平台之后，该主机默认使用本地磁盘介质作为存储，其它主机不能使用。在数据中心中，很多用户选择使用共享存储来承载虚拟机及其数据，目前，H3Cloud云计算管理平台支持IPSAN和FCSAN等类型的存储。采用共享存储的好处是：共享存储往往比本地存储提供更好的I/O性能（尤其在多虚拟机环境下）。H3Cloud云计算管理平台中的高可用性功能需要共享存储作为先决条件，例如HA和动态资源调整等。H3Cloud虚拟机文件系统是一种优化后的高性能集群文件系统，允许多个云计算节点同时访问同一虚拟机存储。由于虚拟架构系统中的虚拟机实际上是被封装成了一个档案文件和若干相关环境配置3文件，通过将这些文件放在SAN存储阵列上的文件系统中，可以让不同服务器上的虚拟机都可以访问到该文件，从而消除了单点故障。系统支持一台故障后，快速切换到另一台的功能，切换时间大概在0-10分钟以内。共享存储服务器集群主存储交换机备用存储交换机RAID备份IP/FCSANVMVMVMVMVMVMVMVMVMVMVMVMVMVMVMVMVMVM图3共享存储组网拓扑结构2.1.2H3CloudvMigrationH3CloudvMigration是实现动态、自动化并自我优化的数据中心所必需的关键特性。H3CloudvMigration支持在物理服务器之间实时迁移正在运行中的虚拟机，利用虚拟机实时迁移技术，不仅在执行硬件维护时无需安排宕机和中断业务操作，而且可以使资源池内的虚拟机不断自动优化，最大限度地提高硬件利用率、灵活性和可用性。x86ArchitectureH3Kernelx86ArchitectureH3Kernel操作系统应用程序vMigration操作系统应用程序操作系统应用程序操作系统应用程序操作系统应用程序操作系统应用程序操作系统应用程序操作系统应用程序操作系统应用程序操作系统应用程序操作系统应用程序共享存储图4虚拟机迁移技术利用H3CloudvMigration在物理服务器之间实时迁移虚拟机的过程是通过两项底层技术实现的。(1)虚拟机操作系统和应用程序的完整状态被封装在共享存储设备上的一组文件之中，这样的共享存储设备可以是光纤通道（FC）、iSCSI存储区域网络（SAN）或网络连接存储（NAS）。通过H3Cloud高性能集群文件系统，多个物理服务器主机可以同时访问同一个虚拟机文件。4(2)虚拟机的内存映像和精确执行状态可通过高速网络在各物理服务器主机之间迅速传递。H3CloudvMigration通过在一个位图中保存对现行内存处理过程的跟踪，使用户在数据传递期间察觉不到性能变化。一旦整个内存和系统状态被复制到目标物理服务器主机之后，H3CloudvMigration就会中止源虚拟机的运行，随后，将位图复制到目标物理服务器，并在目标物理服务器上继续运