微软集群知识讲座.ppt

《微软集群知识讲座.ppt》由会员分享，可在线阅读，更多相关《微软集群知识讲座.ppt（71页珍藏版）》请在三一办公上搜索。

1、1,使用服务器群集实现高可用系统(MSCS),微软（中国）有限公司,2,议程,为什么使用群集?什么是服务器群集?Windows 2000中的服务器群集Windows 2000 特性配置和调试地理分散的群集（Geographically Dispersed Clusters）存储区网络（Storage Area Networks）Windows.Net Server 的新特性支持群集的应用软件Cluster-Aware Application案例分析,3,为什么使用群集?,4,现今服务器的要求,任务:完成传递商业的价值运行要求的应用&服务处理可以预料的负荷提供适当的响应时间可扩充性:与商业一起成

2、长在用户,数据,交易上快速增长在需要时，应付不可预知的波动可用性:按时完成,不让用户等待容忍典型的硬件故障从意外故障重快速恢复通过定期维护减少停机时间支持灾难恢复的工具&技术设计降低成本:与商业价值相当最初投资的费用日常管理的费用维护&升级的费用,5,为什么使用群集?,一个群集是一组计算机共同提供一个应用平台可扩展增加更多的节点，增加 CPU,内存,IO 和网络带宽高可用多个计算机具有独立的故障方式 计划中的冗余和意外宕机管理起来象是一个计算机,扩展性和可靠性,6,扩展性:向上和向外,“Scale Up”Datacenter Class machines“big iron”Cluster fo

3、r 可用性“Scale Out”Commodity serversCluster for 扩展性 and 可用性,7,Windows 2000 群集,当前三种群集技术网络负载均衡-Network Load Balancing目标:32-各节点IP通讯负载均衡，通过一组同样的服务器好处:扩展性,可用性组件负载均衡-Component Load Balancing目标:8 个节点COM组件动态负载均衡，通过一组同样的服务器好处:扩展性,可用性服务器群集-服务器群集目标:支持长时间运行的应用的故障转移好处:可用性,扩展性（对于一些应用）,8,Clients,多种群集在一起工作,*=Component

4、 Load Balancing is delivered through the Application Center 2000 product.,9,什么是服务器群集?,10,什么是服务器群集?(Aka MSCS and Wolfpack),提供故障转移和健康监视支持，对于长时间运行和状态连续的应用应用安装在群集中所有节点上要求共享存储的架构当应用在群集的节点间故障转移时仍能访问相同的数据集所有节点必须在相同的域和IP子网中由群集来监视节点和应用的运行情况节点监视通过群集内的心跳线（heartbeats）应用监视通过应用特殊的“plug-ins”或“resource dlls”所有节点应该至

5、少有2 块网卡1 NIC 用于群集心跳线，1 NIC 用于客户端访问,11,微软群集服务Active-Passive Configuration,Secondary,Data,SCSIShared Disk,VirtualServer,Primary,12,微软群集服务Active-Active Configuration,VirtualServer2,DataPartition1,DataPartition2,Primary VirtualServer1Secondary VirtualServer2,Primary VirtualServer2Secondary VirtualServer

6、1,VirtualServer1,13,微软群集服务Cluster Service in Action,DataPartition1,DataPartition2,Heartbeat,VirtualServer1,VirtualServer2,14,微软群集服务Cluster Service in Action,DataPartition1,DataPartition2,VirtualServer1,VirtualServer2,15,服务器群集 SCSI 配置(2 node only),Client PCs,Public network,Private network(heartbeats,

7、status,control),RAID disk sets,SCSI,Server Cluster Nodes,TIP:Make sure host adapters have different SCSI Ids(Factory default is usually 7),External SCSI Bus Termination,16,服务器群集光纤通道配置,Client PCs,Public Network,Private network,RAIDdisk sets,Server Clusternodes,Redundant fiber channel fabrics,Dual-pat

8、h HBA drivers for fault tolerant access to storage,17,不同的群集视角,网络客户看到的是一个或多个“虚拟的”服务器每个虚拟服务器拥有一个名称,IP 地址每个提供一组服务 e.g.数据库,文件&打印共享每个虚拟服务器由一个群集的节点托管每个节点可以托管多个虚拟服务器如果托管的节点失效,它的虚拟服务器转移到其他的可用的节点上用户端继续使用同样的IP地址和网络名称去访问应用或服务,18,Cluster name,Node name,Node name,Virtualserver name,Virtualserver name,Virtualserv

9、er name,Virtualserver name,服务器群集命名空间物理上命名空间,19,Node 1,Node 2,Virtualserver 1,Virtualserver 2,Virtualserver 3,IP address:Network name:NTSENode1,IP address:Network name:NTSENode2,IP address:Network name:NTSE-VS1,IP address:Network name:NTSE-VS2,IP address:Network name:NTSE-VS3,Internet Information Ser

10、verSQL,MTSMSMQ,Microsoft Exchange,服务器群集命名空间逻辑上命名空间,20,额外的可用性技巧,共同故障的容错RAID disks双重电源,双重风扇,多块网卡,etc.严格的操作程序备份&恢复实践配置标准化&改变控制警报,事变管理运行中的容量监视和计划灾难恢复计划Off-site 数据存储灾难响应计划&排练租借或特有的灾难恢复站点数据管理(e.g.,backups,real-time mirroring)环境管理(e.g.,application licenses)网络管理(e.g.,subnet replication services),21,Windows

11、2000新特性,22,Windows 2000 增强的特性 部署,简化群集部署Simplified setup wizardSetup from with Add/Remove Programs appletScripted,unattended setup(using Unattend.txt)Can use SYSPREP with scripted cluster setupStep-by-step installation guideWindows 2000 Deployment Guide,Chapter 18Rolling upgrade support from NT44-节点群

12、集-Windows 2000 DatacenterSurvive double&triple server lossesLower cost of excess capacity(N+I possible)More workload flexibility for server consolidation更多的操作系统特性支持高可用DHCP,WINS,DFS,NNTP,SMTP,23,Windows 2000增强的特性 群集管理,更容易的群集管理Cluster admin console accessed from Computer Management MMC snap-inApplicat

13、ion Wizard for Simplified Virtual Server ConfigurationCLUSTER command line utilityExpanded online documentation远程&分布式群集管理Administration via Terminal ServicesScripted administration(KB article Q245656)Cluster admin GUI in Windows 2000 AdminpakCLUSTER command for remotable,command-line management,24,高

14、可用性特性Base server reliability improvements Plug-and-play cluster disks Properties to control when chkdsk runs(KB article Q223023)Printing EnhancementsSupport for Crypto key export/import,Windows 2000增强的特性 高可用性,25,应用部署更简单Cluster support requirements&guidelines in Windows 2000 Logo specificationCluster

15、 APIs available to programmers on on every platformCOM 支持(for scripting etc.)MSCLUS COM Interface to CLUSAPI群集日志增强(KB article Q234767)Enabled by default with size control&wraparoundLog entries fully time/date stamped,Windows 2000增强的特性 应用开发和群集调试,26,转动式升级,Upgrade,Resource2,Resource1,Resource2,1,Resour

16、ce1,Resource2,Upgrade,27,Windows 2000部署和调试技巧,28,部署的最佳实践General,只有在 HCL 中被认可的配置Building clusters from parts may not work机器容量必须足以承受所有失败的情况In a 2 node cluster,to survive a single failure with no degradation,each node must be configured to run at 50%Thorough performance testing should be done before dep

17、loyment to ensure that performance degradation will not occur,29,部署的最佳实践Storage,存储架构是许多群集问题的根源共享存储要求不能有任何磁盘与Boot/system/pagefile 在同一条总线上SCSI 只支持 2 节点Ensure all devices have unique Idshost adapters typically ship as Id 7Ensure correct SCSI terminationExternal termination using“Y”connector allows mach

18、ines to be powered down光纤通道Ensure multiple,independent paths to storageMultiple HBAs with multi-path device driversEnsure all HBAs at same firmware revision,30,部署的最佳实践Networks,内部网和外部网At least 2 NICs per machineOne NIC dedicated to private and one NIC for both private and public(PCI NICs only)确保所有相关的

19、网卡都在相同的子网上MUST only be 1 NIC per subnetSwitches/routers 需要支持 arp cache flush protocol可用性-WINS,DNS,DHCP servers必须是静态IP addresses,DHCP 不支持NETBIOS is required for CluAdmin,31,所有节点在相同的域中只能使用域帐号在虚拟服务器的应用使用NTLM验证Kerberos authentication on Windows.Net ServerKerberos will also be available on Windows 2000 S

20、P3(DCR being tested)确保域架构高可用在 Windows 2000 建议群集节点使用成员服务器而不是域控制器(Q281662),部署的最佳实践Security and Domain Infrastructure,32,排错工具,Event ViewerFailures logged in event logEvent log entries are replicated to all nodes in the cluster(lazy replication)Windows.Net Server logs successful group moves,which allows

21、 for auditing of admin operationsCluster LogEnabled by default starting in Windows 2000ClusLog tool available for aggregating multiple logs into a single viewClusDiagNew tool being developed for Windows 2000 and Windows.Net Server to validate cluster configurations,33,Quorum排错,NT 4.0 Q172951 How to

22、Recover from a Corrupted Quorum LogStart using/noquorumlogging optionRun chkdskDelete MSCSQuolog.log,MSCS*.tmpStop,restart serviceWindows 2000 Q245762 Recovering from a Lost or Corrupted Quorum LogStart using/resetquorumlog optionIf necessary restore using“ClusRest”toolAvoid stability issues by incr

23、easing log size to 4096KBThis is the default size on Windows.Net Server,34,硬盘替换,NT 4.0 FTEdit.exe and Registry manipulationQ243195 Event ID 1034 for MSCS Shared Disk After Disk ReplacementWindows 2000-Dumpcfg.exe to change disk signatureQ224075 Disk Replacement for Windows 2000 Server ClusterWindows

24、.Net Server ASR,ConfDisk.exeQ287556 How to use ASR to Recover a Server Cluster NodeQ280425 Recovering from an Event ID 1034 on a Server ClusterConfDisk.exe replaces DumpCfg.exe,35,备份和恢复,NT4 Manual Backup/RestoreWindows 2000 Added backup APIsEnabled System State backups.Restores via ClusRest.exeQ2489

25、98 How to Properly Restore Cluster InformationWindows.Net Server Full SupportBackup/Restore full support via NTBackup.exe with the System StateRestore to all nodes in the ClusterAutomatic System Recovery(ASR)to recover a Cluster nodeQ287556 How to use ASR to Recover a Server Cluster NodeQ286422 How

26、to Backup/Restore a Server Cluster under Windows.Net Server,36,动态磁盘,服务器群集不支持动态共享盘Veritas has a Volume Manager for Windows 2000 product that supportsDynamic Disks for Shared DisksSoftware FT Sets on Clustered DisksDynamically change shared disk sizeALL support for clusters running Veritas Volume Mana

27、ger should be referred to Veritas directlyQ237853 Dynamic Disk Configuration Unavailable for Server Cluster Disks,37,Windows 2000地理分散的群集-Geographically Dispersed Clusters(aka multi-site clusters),38,地理分散的群集,服务器群集支持群集的节点可以分布在不同的地理位置Limited to 2 site deployments in initial releaseWindows 2000 Advanced

28、 Server and Datacenter Server2 node AS/2 and 4 node DatacenterNew HCL categories specific to multi-site configurationsCertified configuration includes hardware plus software一些警告.,39,the good,bad and ugly,所有群集节点必须在相同的子网上Requires VLAN-like technology across sites站点之间的网络反应时间必须保证小于 500msLatency tool ava

29、ilable to qualify inter-site links when deployed at customer sites存储数据必须在站点间镜像 Storage solutions provided by different vendors(OEMs,IHV and ISVs)Quorum disk MUST be synchronously replicated across the sitesData disks SHOULD be synchronously replicated to ensure cross site consistency,40,地理分散的群集,Late

30、ncy=500 MS,Virtual LANs Appear as 1 Network to Cluster,Minneapolis,VLAN,Chicago,41,Site A Site B,Up to 100km,Path A,Path B,FC switch,FC switch,FC switch,FC switch,one way,地理分散服务器群集,SAN1,SAN2,R/W,R,R/W,R,VLAN,42,Windows 2000Storage Area Network Deployments,43,群集和 SAN 支持,多个群集可配属到相同的存储结构上存储设备必须在Multi-c

31、luster device HCLServer+Fabric+Storage 必须是作为一个完成的方案在群集的HCL目前没有针对fabric switches认证过程,Windows 2000服务器群集支持存储区网络,Cluster,Cluster,Cluster,JBOD,StorageArray,SAN Fabric,44,配置 SANs,使用区域划分隔离每个群集Put cluster nodes and cluster shared devices into cluster specific zones Disk arbitration uses“bus reset”,Cluster 3

32、,Cluster 2,Cluster 1,JBOD,Cluster 1Zone,Cluster 2Zone,Cluster 3Zone,StorageArray,45,Windows.Net ServerNew Features coming soon,46,Windows.Net Server Highlights,群集配置向导Simplified cluster setup and configuration“Generic Script”resource typeNew resource type to allow existing applications to be made clu

33、ster-aware using VBScript or JScript更大的群集4 nodes in Advanced Server8 nodes in DatacenterWMI 支持管理和事件Active Directory 集成End-to-end Kerberos authentication and delegation,47,额外的特性概要(1 of 2),Rolling upgrade NT4 to Windows 2000 to Windows XPCluster account password utilityChange cluster service account p

34、assword without taking down the clusterEnhanced SAN supportTargeted reset of storage devices rather than full bus reset-requires supported by HBA device drivers“Single storage bus”supportFile system enhancementsPhysical disk resources no longer require drive lettersMountpoints supported in a cluster

35、(support 23 disks)Client-side caching for clustered SMB sharesMultiple DFS roots in a cluster,48,额外的特性概要(2 of 2),Printer drivers replicated to all cluster nodesSimplified configuration of MSDTCCreate an MSDTC resource no more COMCLUSTAutomatic Server Recovery(ASR)supportRestores disk signatures for

36、all disk(including shared disks)Rebuilds a single node including cluster registry stateCan be used to rebuild a complete cluster(including quorum disk)Terminal Server and 服务器群集 can co-exist on same nodes(Note:NO failover of TS sessions)“Majority Node Set”quorum resourceResource Dll enhancementsSuppo

37、rt for rolling upgrade without shutting down the clusterNotification of terminate or offline reasonSupport for group affinity,49,群集拓扑结构,4 nodes in Advanced Server8 nodes in DatacenterFully functional cluster support for ia64,50,3 步群集配置,Analyzing the configuration,Show the proposedconfiguration,Make

38、the changes,51,支持群集的应用软件Cluster-Aware Application,52,A Failover Cluster 2 Node,Virtual SQL Server(s),Private Network,Public Network(s),Node 1,Node 2,53,A Failover Cluster 4 Node,54,SQL中数据分区,Client,User(View),Server A,UsersAF(Table),Server B,UsersGZ(Table),CREATE TABLE UsersAF(UserName VarChar(50)NOT

39、 NULL PRIMARY KEY CHECK(UserName=F),Email VarChar(50),FirstName VarChar(20),LastName VarChar(30),CREATE TABLE UsersGZ(UserName VarChar(50)NOT NULL PRIMARY KEY CHECK(UserName=G),Email VarChar(50),FirstName VarChar(20),LastName VarChar(30),CREATE VIEW Users ASUNION ALL,CREATE VIEW Users ASUNION ALL,DT

40、C needs to be running on both machines Need to set XACT_ABORT on(execute SET XACT_ABORT ON),Read or Update,55,Exchange 5.5 群集,Active/passiveNot cluster-AwareGeneric Resource DLLLow ROI,56,Exchange 2000群集,Active/ActiveVirtual Server conceptNodes supported:2 Nodes RTM4 Nodes later with Data Center4 St

41、orage Group limitation,57,Exchange Cluster Groups,Exchange Virtual Server(EVS)Multiple EVSs per nodeFinest granularity for failover1 or more SGs,Exchange Virtual Server,IP AddressNetwork NameShared Storage Exchange Resources,58,Active/Active Exchange,MAPI,EVS1,StorageGroup 1,Name 1,Name 2,Name 3,Sto

42、re,IIS,59,Client Experience,EVS2,HTTP,POP,IMAP,MAPI,60,Exchange群集,Exchange 5.5 vs.Exchange 2000Cluster APIsActive/Passive vs.Active/ActiveExchange 2000 Configuration Support2-Node A/A:(1VS,2SG,1000 Connections)2-Node N+1(1VS,4SG)4-Node N+1(1VS,4SG)ROI is possible with N+1,61,Exchange中的数据分区,Front-end

43、 server farm,Front-end servers load balanced using Network Load Balancing.Content replication and state managed separately,Exchange Mailbox servers clustered using Server clustering.Redirection to the correct Virtual server hosting the mailbox is done by querying AD.,Public Folders,Application Serve

44、rs,Web Client,Domain Controller,PFServer1,ASServer1,DC1,Firewall 1,Internet,Firewall 2,62,Case-study,63,海关总部,2 节点群集 Windows2000 Advanced Server and SQL Server 2000 Enterprise 数据收集和分析平台,64,中国移动,4-节点Exchange 2000 企业版统一消息平台,65,更多资源,Windows 2000 Server FamilyCluster Services&Load Balancing Windows 2000

45、reliability&availability improvements Windows 2000 Server(especially Chapter 18 of Deployment Planning Guide)Windows 2000 Advanced Server online helpMicrosoft Support Centerhttp:/,66,更多资源,Certified Cluster Hardware Configurations:Choose“Cluster”hclMicrosoft TechNet:High Availability technet/avail/MSDN Online:Platform SDK Components for Windows Base ServicesMicrosoft research,Jim Grays web site/Gray/Microsoft Cluster Server Troubleshooting and Maintenance White Paper(NT 4.0)library/techart/mscstswp.htm,67,68,69,70,71,