方案-數據中心的業(yè)務連續(xù)性規(guī)劃與設計

1、業(yè)務永續(xù)數據中心的業(yè)務連續(xù)性規(guī)劃與設計客戶需求這個世界比過去有著更多的風險Financial TimesDisaster recovery: The crucial thing is to be prepared1USA TODAYTheft of personal data more than triples this year2The Economic TimesData backup, recovery becoming critical to all3環(huán)境在不斷變化風險接觸范圍在不斷擴大全球及區(qū)域間的依存關系在不斷增大供應鏈每時每刻都存在中斷的風險業(yè)務中斷將導致更大的影響宕機可能導致更

2、大的財務影響宕機可能對品牌造成傷害宕機可能導致數據失去完整性更繁雜的規(guī)范產業(yè)和監(jiān)管標準在不斷變化產業(yè)分工在地理分布上更趨分散每個國家都可能有自己相應的規(guī)范更多的災害經濟危機、恐怖主義、颶風、地震、 停電、火災和疾病的大規(guī)模威脅災難的分類 每年的發(fā)生頻率每次發(fā)生的結果 (單次發(fā)生損失) （美元）1,0001001011/101/1001/1,0001/10,0001/100,0001美元10美元100美元1千美元1萬美元10萬美元100萬美元1000萬美元1億美元病毒蠕蟲磁盤故障組件故障電源故障常見不常見低高自然災害應用中斷數據損壞網絡問題建筑火災恐怖行動/國內動蕩與可用性相關的與恢復相關的業(yè)務

3、連續(xù)運營業(yè)務連續(xù)性問題和挑戰(zhàn)差距不斷加大更多的業(yè)務在線更多的應用和數據增長的需要通過傳統(tǒng)的備份與恢復滿足業(yè)務需求的能力更多復雜的系統(tǒng)更少的恢復時間窗口更小的對停機時間的容忍度日益增加的信息不可用所造成的影響備份與恢復 vs. HA高可用重新運行批處理日終作業(yè)手動的應用與數據的恢復丟失數據最好的意圖 vs. RTO, RPO, SLA規(guī)范的設計收入和利潤受損失 負面的社會影響罰款和罰金涉及法律依從及會計的問題員工的勞動強度和費用對日常的業(yè)務規(guī)劃和運作產生影響 60%的客戶正在關注如何提高可用性 接近50%的客戶希望有顯著的安全提升 超過25%的客戶希望實施高可用集群業(yè)務延續(xù)運性之考慮Consid

4、eration for Business ContinuityFault-tolerant hardware, redundancy, automatic detection and isolation, predictive analysis, call-homeReal-time replication of data over metropolitan and/or continental distancesAutomated protection against unplanned outages with meeting recovery point and recovery tim

5、e objectives高可靠性High Availability數據復制Data Replication災備技術Disaster Recovery業(yè)務延續(xù)運性Business ContinuityIBM Power Systems高可用性解決方案 HA和DR 的差別High Availability自動的接管一般適用發(fā)生在本地的錯誤針對物理設備的保護服務器硬盤適配器卡網絡針對致命的軟件錯誤的保護操作系統(tǒng)數據庫應用服務Disaster Recovery手動的切換流程針對于主站點完全失效方面的保護覆蓋的錯誤包括：HA方案失效主站點 (基礎架構)失效邏輯錯誤(如應用或數據)致命的用戶失誤導致的原因

6、是自然災害、戰(zhàn)爭、對主站點有嚴重的影響災難復原計劃是必要的基本 可用性無 數據丟失恢復點目標Data CurrencyLatest持續(xù)可用性數據傳輸 （每個交易的價值）可用水平恢復時間目標和縮短的計劃內停機時間Availability LevelSAN 磁盤備份服務 多服務器解決方案iSeries single-server單服務器solutions解決方案備份周- 日場外存儲RAID-5日志組合磁盤鏡像 SANAIX, Linux, Intel 群集連續(xù)數據復制群集可切換集群 在線維護CUoD高速磁帶LPARTSMBCRSSWA網絡分配器冗余網絡(LAN/SAN)支持業(yè)務連續(xù)性與災備的系統(tǒng)組

7、件面向開放平臺的高可用(High Availability)解決方案的架構Availability by 應用按照高可用的要求來設計應用架構Availability by 中間件DB2 HADR、WAS 集群, CICS 集群Oracle RACAvailability by 操作系統(tǒng)AIX LVM 鏡像、HACMP for AIXAvailability by 硬件冗余服務器冗余的處理器 / I/O適配器卡/ 電源 /內置磁盤RAID技術保護外置磁盤, I/O 總線、SAN 交換機、LAN、LAN 交換機冗余的部件磁盤 RAID多路徑(Multi-Path)軟件（SDD、RDAC)通過磁盤復

8、制的可用性FlashCopy, Metro/Global mirror網絡HardwarePower Systems ( RAS )Live Partition Mobility PowerSystems SoftwarePowerHA PowerHA/XDApplicationOperating SystemAIX, ILive Application MobilityIBM Power Systems High Availability SolutionImpact of Maint on AvailabilityIT Availability 24 x 7 Time for Planne

9、d MaintenanceIs this you? Your users demand continuous availability (24x7)Do you agree? As IT availability approaches 24x7, top-notch maintenance practices become more criticalIs this your problem? As IT availability approaches 24x7, the time for maintenance work approaches zero!The Power Systems Hi

10、gh Availability Solution can show you how POWER6 processors and AIX 6 help to address the maintenance crunch! Learn how Live Partition Mobility, Live Application Mobility, Workload Partitions and PowerHA can enable non-disruptive maintenance anytime! Do you want more availability and less work on we

11、ekends?High Availability Hardware - Reliability, Availability and ServiceabilityIBM Power Systems RAS架構Processor Instruction Retry Alternate Processor Recovery First Failure Data CaptureDDR Chipkill memoryBit-steering/redundant memoryService Processor Failover*Dynamic Firmware Maintenance*Hot I/O Dr

12、awer Add*I/O error handling extended beyond base PCI adapterECC extended to inter-chip connections for the fabric/processor busesMemory and L3 Cache soft scrubbingHardware AssistedL2 & L3 Cache Line DeleteHardware Assisted Memory ScrubbingLive Partition Migration570 Concurrent Add & Cold Repair Prim

13、ary POWER RAS FeaturesHMC required to enable these functions.Primary POWER RAS Features - ContinuedHMC required to enable these functions.Redundant power, fansDynamic Processor DeallocationDynamic processor sparing ECC memoryPersistent memory deallocationHot-plug PCI slots, fans, powerInternal light

14、 path diagnosticsHot-swappable disk baysCore System DesignHigh quality partsFewer parts = Fewer failuresDesigned for low power consumption (less heat = fewer failures)Manufacturing methods, packaging, coolingContinuous System and Commodity Quality ActionsIntegrated RAS featuresFailure Avoidance Meth

15、odologyDesigned for Ease of ServiceFault ResilienceN+1 Power Supplies, regulators, power cordsDual redundant fansDynamic Processor Deallocation and sparingChipkill TechnologyPredictive Failure AnalysisAuto Path Reassignment - data paths, powerProcessor Instruction RetryFault Isolation & Diagnosis Fi

16、rst Failure Data CaptureRun Time Self DiagnosticsService ProcessorRifle-shot repairs (no plug and pray parts replacement approach)System RestoreDeferred RepairConcurrent RepairLED Service IdentificationService ConsolesMigration to Guided MaintenanceSummary of key Power Systems RAS featuresWorld-clas

17、s Hardware RASHigh Availability Hardware - Live Partition MobilityLive Partition Mobility with POWER6*Allows migration of a running LPAR to another physical server Reduce impact of planned outages Relocate workloads to enable growth Provision new technology with no disruption to service Save energy

18、by moving workloads off underutilized serversMovement toa different server withno loss of serviceVirtualized SAN and Network Infrastructure* All statements regarding IBM future directions and intent are subject to change or withdrawal without notice and represent goals and objectives only. Any relia

19、nce on these Statements of General Direction is at the relying partys sole risk and will not create liability or obligation for IBM. Continuous Application AvailabilityWith Live Partition Mobility and Live Application Mobility, planned outages for hardware and firmware maintenance and upgrades can b

20、e a thing of the pastRelocate all partitions from one server to another when performing maintenance. Move the partitions back when maintenance is complete* All statements regarding IBM future directions and intent are subject to change or withdrawal without notice and represent goals and objectives

21、only. Any reliance on these Statements of General Direction is at the relying partys sole risk and will not create liability or obligation for IBM. Workload Balancing with Live Partition Mobility*As computing needs spike, redistribute workloads onto multiple physical servers without service interrup

22、tionAs one server gets overtaxed from a spike in demand, relocate partitions to other servers* All statements regarding IBM future directions and intent are subject to change or withdrawal without notice and represent goals and objectives only. Any reliance on these Statements of General Direction i

23、s at the relying partys sole risk and will not create liability or obligation for IBM. High Availability Operating System - AIXUNIX Reliability, Availability and Serviceability The “Number One” Customer RequirementCompetitionAIX - 2007AIX - 2006AIX 2005Enterprise Continuous Availability CapabilityTi

24、me AIX FunctionalityKernel Storage KeysConcurrent AIX updatesCross System Workload MobilityDynamic Tracing with probevueFunctional Recovery RoutinesComponent TraceMemory Overlay ProtectionParallel DumpLightweight Malloc debugLightweight Memory TraceConsistency CheckersComponent RAS infrastructureAIX

25、 errorlogSubsystem Resource ControllerExploitation of a POWER6 processor hardware feature to provide additional isolation of kernel and application dataStorage keys can prevent invalid changes to memory cause by programming errorsApplication use of POWER6 storage keys is enabled in AIX V5.3 AIX Kern

26、el exploitation of POWER6 storage keys is included in AIX V6.1What is it?AIX exclusive feature not available in UNIX, Linux, or Windows!AIX Storage Keys AIX 6 Concurrent Maintenance Kernel SpaceUser SpaceInterim FixConcurrent updatevmmove() patchemgrvmmove()getgidx()sleepx()Non-disruptive fixes to e

27、xecutable code in a running AIX kernelBase AIX Kernel (/unix), kernel extension, or device driverNo downtime (reboot) required to apply fix and make it activeConcurrent updates will be packaged as Interim FixesFix selected AIX kernel problems without a service outagevmmove()AIX 6 dynamic tracing wit

28、h probevueTrace existing programs without recompiling Dynamic placement of trace probesFor debugging and performance analysisTracable Calls:AIX system calls, application functions, and application calls to library functionsDynamic tracing language called VueInitial support only for “C” programs#!/us

29、r/bin/probevue/* countreads.v */syscall.$1.read.entry count+;interval.*.clock.100 printf(“Number of reads = %dn”, count); count = 0;# countreads.v 404Number of reads = 22Number of reads = 0Number of reads = 1Number of reads = 17.Formatted I/OUserKernelProbe LocationUser Process CodeSome thread hits

30、probe point (1)Branches to probe code (2)Probe code(3)Returns to probe point (4)Thread continues execution(5)Trace ConsumerTrace FileorTrace OutputTrace BuffersE-code“Vue” probe code exampleThe AIX answer to Solaris dtraceThis information is intended only for IBM sellers and Business PartnersAIX V6.

31、1 Workload Partitions (WPAR)Virtualized AIX operating system environments within a single AIX imageEach WPAR shares the single AIX operating system but can be separately managed Applications and users inside a WPAR cannot affect resources outside the WPAREach WPAR can have a regulated share of proce

32、ssor, memory and other resourcesTwo types of WPARSystem WPARs have separate security and appear like a completely separate OSApplication WPARs are manageability wrappers around a single applicationWhat is it?This information is intended only for IBM sellers and Business PartnersAIX V6.1 Live Applica

33、tion MobilityThe capability to relocate a running Workload Partition from one system to another without restarting the applicationThe application running inside the WPAR resumes running after the relocation is complete Works with systems based on POWER4, POWER5 and POWER6 processorsRequires the IBM

34、Workload Partitions Manager for AIXManual or automatic, policy based relocationWhat is it?操作系統(tǒng)停機時間調查: AIX是業(yè)界最穩(wěn)定的操作系統(tǒng)(27個國家400個用戶)The Yankee Group “2007-2008 Global Server Operating Systems Reliability Survey” as quoted in “Windows Server: The New King of Downtime” by Mark Joseph Edwards at /article/

35、articleid/98475/windows-server-the-new-king-of-downtime.html, March 5, 2008 and in /stu/Yankee-Group-2007-2008-Server-Reliability.pdfWin2000Win2003RHELSolarisHP-UXSUSEAIXWe are here!This information is intended only for IBM sellers and Business PartnersAccording to a recent Yankee Group study* of 40

36、0 Windows, Linux and UNIX users, AIX was the most reliable server operating system:“IBMs AIX achieved the highest level of reliability, with corporate enterprises reporting an average of only 36 minutes of downtime per server in a 12-month period”* Source: “Unix, Linux Uptime and Reliability Increas

37、e; Patch Management Woes Plague Windows” 2008 Yankee Group Research, Inc. All rights reservedAIX is “Most Reliable”High Availability System Software - PowerHA- PowerHA/XDIBM PowerHAPowerHA for AIXPowerHA Cluster Management Monitors, detects and reacts to eventsEstablishes a heartbeat between the sys

38、temsEnables automatic switch-overIBM shared storage clustering Can enable near-continuous application serviceHelps eliminate impact of planned & unplanned outagesEase of use for HA operationsPowerHA managing integrated IBM data resiliencyLogical Volume Manager (LVM) Shared switchable disk topologyXD

39、 (optional feature of PowerHA)GLVM (Global LVM) AIX based replication over IPMetro Mirror IBM storage based synchronous mirroringSVCIBM DS8000Smart Assists Application deployment and configuration34PowerHA for AIX V5.5PowerHA V5.5 FeaturesSimplified Management Manage multiple clusters from a single

40、graphical user interfaceCan run on a server outside of the cluster Support for TCP/V6 connections to clients New focus on IPV6 from US governmentPowerHA/XD V5.5 Disaster Recovery Global Logical Volume Manager*Global Logical Volume Manager (GLVM) asynchronous mode mirroringAsynchronous mode enables g

41、eographic dispersion San Volume Controller Global MirrorAsynchronous replication for geographic dispersion*GLVM Asynchronous mode generally available March 2009Shared storage clustering TopologyNetwork ClientsSerial HeartbeatPower Cluster NodePower Cluster NodeIP NetworkService & Standby Network Ada

42、ptersShared DiskIP Heartbeats主機主機磁盤1磁盤2Switched Disk Cluster (Local only)本地存儲雙機LVM基于AIX功能（軟件免費）完全冗余，無切換中斷時間特別適合24X7環(huán)境存儲可靠性幾何級提高雙存儲可輪流定期修整維護PowerHA/XD (HACMP/XD) 延伸PowerHA的概念到更遠的距離利用 SVC or DS8000/DS6000/ESS鏡像技術RouterRouterDS8/6/ESS MirroringPrimaryESS/DSSecondaryESS/DS生產站點恢復站點SVCSVCSVC Mirroring orG

43、LVM Mirroring利用 Global Logical Volume Manager (GLVM) 技術 IBM AIX Multi System Data ResiliencyPowerHA for AIXStrategic building block for IBM AIX High Availability and Disaster Recovery solutionsIntegrated and optimized with IBM AIX Cluster ResourcesSwitched DiskStorage agnosticLVM mirrored copy of da

44、ta HA (Local only)Switched Disk ClusterGeographic Logical Volume Manager IP deployed mirrorStorage agnosticHA and DRPowerHA XDMetro MirrorMetro MirrorDS8000 & SVCIBM FlashCopyHA and DR FlashCopyGeographic MirroringGLVM ClusterMetro Mirror/Global Mirror ClusterBasic San Copy ServicesMetrol MirrorBoot

45、 From SAN DR/Tape BackupDR onlyFlashCopyGlobal Mirror高可用性整體解決方案數據庫服務器應用服務器RAID 5 或RAID 10雙數據拷貝冗余SAN 網絡服務器集群并行數據庫冗余 網絡應用伸縮性邊緣設備高可用性的實現層次高性能高可靠性的并行文件系統(tǒng) - GPFS什么是GPFS集群: 可以擴展至4096節(jié)點，高速、穩(wěn)定地通訊，單點管理與控制；共享磁盤：可以從集群中的任一節(jié)點直接訪問磁盤上的數據；并行訪問：所有節(jié)點訪問所有磁盤的數據流并行實現；IBM為AIX和Linux集群系統(tǒng)設計的共享磁盤的并行文件系統(tǒng)為什么要用GPFS并行文件系統(tǒng)應用需求:多個

46、節(jié)點訪問同一個數據文件或數據庫高性能文件訪問故障恢復文件系統(tǒng)需求:可訪問：從任一節(jié)點訪問所有文件；動態(tài)擴展：能動態(tài)地增加或減少節(jié)點與存儲；文件唯一存在：使得在集群環(huán)境中的應用開發(fā)更加容易；高容量：TB級文件，PB級的文件系統(tǒng)，測試過2PB；高吞吐率：單文件的訪問可達GB/s，現最高記錄為102GB/s；數據并行訪問：并行訪問單個文件或多個文件；可靠和容錯：當某個節(jié)點、磁盤或連接出現問題時，仍然可以提供服務；GPFS的主要優(yōu)勢高性能條帶化文件讀寫提高并發(fā)訪問性能，實測帶寬可達數百GB智能預取機制和客戶端數據緩存機制降低讀寫延遲分布式的元數據服務器和字節(jié)鎖管理可自定義數據塊大小可，從16K到4MN

47、SD支持InfiniBand RDMA高可用性仲裁管理和自動故障切換支持多路徑磁盤訪問，每塊邏輯盤可支持8個NSD Server支持元數據和用戶數據的復制功能在不停止服務的情況下可以動態(tài)加入和移除節(jié)點或磁盤，支持在線升級支持日志功能，實現系統(tǒng)快速恢復高可擴展性支持最大299 字節(jié)的文件系統(tǒng)和20億個文件支持數千個節(jié)點的集群系統(tǒng)支持不同存儲、網絡、處理器和操作系統(tǒng)易管理自動在各個節(jié)點間同步配置文件和系統(tǒng)信息可在集群內任何一個節(jié)點上完成對GPFS的管理任務，命令將在所有節(jié)點上生效管理網絡和數據網絡可以分開其他支持信息生命周期管理支持CNFS支持快照功能和數據備份提供DMAPIGPFS的兩種基本配置SAN 組織方式NSD 組織方式SANI/O ServersLANNSD ClientsSANGPFS Nodes Database ApplicationsDB2Oracle RACSAPGrid ApplicationsScientific ComputingLife SciencesAnalyticsWeb ApplicationsEmail servicesWeb Server FarmOnline Data Storage Digital MediaAnimationBroadcastingVideo SurveillanceHighly Avai