VMware虛擬化最佳實踐及規(guī)劃

虛擬化最佳實踐及規(guī)劃議程應用實施范圍考慮服務(wù)器采購考慮虛擬機部署考慮管理維護考慮議程應用實施范圍考慮服務(wù)器采購考慮虛擬機部署考慮管理維護考慮應用實施范圍總體原則不適合采用虛擬化的應用具有特殊硬件訪問要求的應用高性能圖形顯卡

不適用虛擬化

特殊的串/并行加密設(shè)備不適用虛擬化USB設(shè)備連接需求

可能不適用，可采用外置USB設(shè)備代替，需經(jīng)過測試即使在高配置的服務(wù)器上仍然具有很高負載的應用

可能不適用，需分析當前服務(wù)器配置情況可以采用虛擬化的應用除上述不適合采用虛擬化的應用之外的所有應用可根據(jù)應用遷移的復雜程度決定虛擬化先后順序較易實現(xiàn)P2V的應用可先做遷移，如可用Converter工具直接遷移的應用較難或不能做P2V遷移的應用可考慮采用重新安裝方式后遷根據(jù)管理的需要決定是否做虛擬化虛擬化轉(zhuǎn)變過程對現(xiàn)有業(yè)務(wù)的影響程度轉(zhuǎn)變?yōu)樘摂M化后對現(xiàn)有管理的影響程度部門之間協(xié)調(diào)的難易程度虛擬化宿主服務(wù)器的部署類型垂直擴展與水平擴展部署模式不同資源池的“量子化”模型物理主機HypervisorVMOSAppVMOSAppVMOSAppVMOSAppVMOSAppVMOSAppVMOSAppVMOSAppVMOSAppPhysicalHostHypervisorVMOSAppVMOSAppVMOSAppPhysicalHostHypervisorVMOSAppVMOSAppVMOSApp物理主機HypervisorVMOSAppVMOSAppVMOSApp不同資源池的類型垂直擴展的主機模式提供更大的連續(xù)性資源空間更容易滿足不同負載的吻合性要求可提供更高的資源利用率水平擴展的集群主機模式更像是一組小池子的集合多組小容量資源池需要更多的監(jiān)控管理模塊化使用既有優(yōu)點也有缺點影響虛擬化部署的參數(shù)體系功能的多樣化服務(wù)器的重要性獨立服務(wù)器不重要重要(獨立的,本地存儲等等)(群集的,多主機的等等)服務(wù)器農(nóng)場(水平擴展服務(wù)器群集，公用服務(wù)器等等)(后端辦公,本地應用等等)負載約束技術(shù)約束商業(yè)約束困難度增加負載約束不同的資源組都需要分別考慮CPU利用率磁盤I/O網(wǎng)絡(luò)I/O內(nèi)存利用率虛擬化造成的額外負載通常也要做一定考慮，如磁盤和網(wǎng)絡(luò)的I/O會增加CPU的負擔iSCSI存儲訪問也會增加CPU負擔運維的周期性負載變化也必須考慮進來月末負載變化年末負載變化技術(shù)約束技術(shù)約束通常主要是指：兼容性（指系統(tǒng)/應用的兼容性）關(guān)聯(lián)性（如系統(tǒng)是摩格邏輯組的一部分）大部分環(huán)境下，這些約束包括了：網(wǎng)絡(luò)連接（子網(wǎng)段級別）應用之間的互連性相關(guān)的存儲使用技術(shù)所用的硬件和外設(shè)軟件支持度和認證這些約束條件根據(jù)虛擬化在內(nèi)核上下實現(xiàn)的不同而有所不同共享與分離OS鏡像模式商業(yè)和流程的約束規(guī)模較小或集中的情況下容易被忽略的約束在實驗室的測試環(huán)境可以不考慮，但生產(chǎn)環(huán)境必須要考慮在虛擬化中常見的商業(yè)和流程約束包括：維護窗口和凍結(jié)改變地理位置和其他物理限制運維環(huán)境，安全區(qū)域，應用分層部署考慮商業(yè)組織，部門以及客戶法規(guī)政策的考慮與限制忽視這些約束條件將可能導致不可預知的結(jié)果具體情況具體分析，根據(jù)目標制定計劃議程應用實施范圍圍考慮服務(wù)器采購考考慮虛擬機部署考考慮管理維護考慮慮虛擬化中使用用的硬件應滿滿足兼容性列列表要求ESX服務(wù)器硬件配配置考慮要點點–CPUsESX調(diào)度CPU周期滿足虛擬擬機和ServiceConsole的處理請求可用的CPU目標數(shù)量越多多，ESX管理這個調(diào)度度機制的效果果越好（單臺服務(wù)器器配置8個以上的CPU核會有最好的的效果）超線程技術(shù)并并不能提供等等同于多核處處理器的好處處；建議關(guān)閉閉CPU的超線程功能能（如果有的的話）使用具有EM64T能力的IntelVT或AMDV技術(shù)的CPU可以同時支持持運行32位和64位的虛擬機采用同一廠商商、同一產(chǎn)品品家族和同一一代處理器的的服務(wù)器組成成的集群，可可以獲得最好好的VMotion兼容能力ES的EnhancedVMotion兼容性擴大了了原有VMotion的兼容能力-《AlleviatingConstraintswithResourcePoolsLiveMigrationwithEnhancedVMotion》參見《BestPracticesforSuccessfulVIDesign》》ESX服務(wù)器硬件配配置考慮要點點-內(nèi)存內(nèi)存資源往往往比CPU資源更會成為為潛在的瓶頸頸在某些時候，，虛機環(huán)境的的內(nèi)存使用量量可能會超過過物理內(nèi)存值值：Hostswapfile(盡量少用以獲獲得最佳性能能)TransparentPageSharing（多個虛機共共享相同內(nèi)存存頁面）注意服務(wù)器特特定的內(nèi)存配配置要求DIMMsizes,bankpairing,parity,upgradeconsiderations(mixandmatchorforkliftreplacement)盡可能將服務(wù)務(wù)器配置到最最大內(nèi)存，采采用最大容量量的內(nèi)存條（（特別是當沒沒有配滿全部部內(nèi)存條時））參見《BestPracticesforSuccessfulVIDesign》》虛擬架構(gòu)的基基本網(wǎng)絡(luò)連接接部件組成(Managementvirtualmachine)(Vmotion,iSCSI,NFS)(VMconnectivity)PortGroupPortGroupPortGroupESX服務(wù)器硬件配配置考慮要點點-網(wǎng)絡(luò)參見《BestPracticesforSuccessfulVIDesign》》最少配置一個個虛擬交換機機，測試環(huán)境境可用2個虛擬交換機機，生產(chǎn)環(huán)境境建議最少配配置3個虛擬交換機機虛擬交換機可可同時支持3種類型的端口口組(ServiceConsole,VMkernel,VM)建議將ServiceConsole、VMkernel和虛機端口組組各自使用自自己的虛擬交交換機可用VLAN技術(shù)分割不同同的端口組對于使用VMotion和DRS功能的服務(wù)器器集群，網(wǎng)絡(luò)絡(luò)配置應該相相匹配（虛擬交換機機的數(shù)量與網(wǎng)網(wǎng)絡(luò)卷標名應應保持一致））ESX服務(wù)器ServiceConsole使用固定IP，配置正確的的speed和duplex。ESX服務(wù)器硬件配配置考慮要點點-網(wǎng)絡(luò)-虛擬交換機和和端口組參見《BestPracticesforSuccessfulVIDesign》》ESX服務(wù)器，虛擬擬交換機，物物理網(wǎng)卡出于冗余的考考慮，每個虛虛擬交換機建建議至少分配配兩個物理網(wǎng)網(wǎng)卡每個ESX服務(wù)器的物理理網(wǎng)卡/口數(shù)量取決于于準備配置的的虛擬交換機機的數(shù)量如果3種類型的端口口組（SC,VMkernel,VM）都在不同的的虛擬交換機機上，生產(chǎn)環(huán)環(huán)境建議至少少6個物理網(wǎng)卡/口如果給包含虛虛擬機端口組組的虛擬交換換機分配更多多的物理網(wǎng)卡卡/口，可以獲得得負載均衡的的好處ESX服務(wù)器硬件配配置考慮要點點-網(wǎng)絡(luò)基本組件件參見《BestPracticesforSuccessfulVIDesign》》物理網(wǎng)卡/口與物理交換換機同一個虛擬交交換機上的不不同物理網(wǎng)卡卡/口應連接到不不同的物理交交換機上將一個集群中中所有服務(wù)器器的VMotion功能端口組所所使用的物理理網(wǎng)卡/口都連到同一一套物理交換換機上（同樣遵循上上述第一條規(guī)規(guī)則）ESX服務(wù)器硬件配配置考慮要點點-與物理網(wǎng)絡(luò)的的連接參見《BestPracticesforSuccessfulVIDesign》》Example1:BladeServerwith2NICPortsvSwitchvmnic0SCvmkernelActiveStandbyvmnic1CandidateDesign:TeambothNICportsCreateonevirtualswitchCreatethreeportgroups:UseActive/StandbypolicyforeachportgroupPortgroup1:ServiceConsole(SC)Portgroup2:VMotionPortgroup3:VMtrafficUseVLANtrunkingTrunkVLANs10,20,30oneachuplinkPortgroup1VLAN10Portgroup3

VLAN30Portgroup2

VLAN20VLANTrunks(VLANs10,20,30)Example2:Serverwith4NICPortsvmnic0SCvmkernelActiveStandbyvmnic1CandidateDesign:CreatetwovirtualswitchesTeamtwoNICstoeachvSwitchvSwitch0(useactive/standbyforeachportgroup):Portgroup1:ServiceConsole(SC)Portgroup2:VMotionvSwitch1(useOriginatingVirtualPortID)Portgroup3:VMtraffic#1Portgroup4:VMtraffic#2UseVLANtrunkingvmnic1andvmnic3:TrunkVLANs10,20vmnic0andvmnic2:TrunkVLANs30,40Portgroup4

VLAN40VLANs10,20vSwitch0Portgroup1

VLAN10Portgroup2

VLAN20vSwitch1vmnic2vmnic3Portgroup3

VLAN30VLANs30,40Example3:Serverwith4NICPorts(SlightVariation)vmnic0SCvmkernelActiveStandbyvmnic1CandidateDesign:CreateonevirtualswitchCreatetwoNICteamsvSwitch0(useactive/standbyforportgroups1&2):Portgroup1:ServiceConsole(SC)Portgroup2:VmotionUseOriginatingVirtualPortIDforPortgroups3&4Portgroup3:VMtraffic#1Portgroup4:VMtraffic#2UseVLANtrunkingvmnic1andvmnic3:TrunkVLANs10,20vmnic0andvmnic2:TrunkVLANs30,40VLANs10,20vSwitch0Portgroup1

VLAN10Portgroup2

VLAN20vmnic2vmnic3Portgroup3

VLAN30VLANs30,40Portgroup4

VLAN40ServerswithMoreNICPortsMorethan4NICPorts—DesignConsiderationsWithTrunks(VLANtagging):UsepreviousapproachandscaleuptomeetadditionalbandwidthandredundancyrequirementsAddNICstoNICteamsupportingVMtrafficVLANTaggingalwaysrecommended,butoptionsifNICsavailable:DedicatedNICforVMotionAtleastoneNICDedicatedNICsforIPStorage(NFSand/oriSCSI)UsuallytwoteamedNICs(considerIP-hashðerchannelifmultipledestinations

andMulti-ChassisEtherchannelemployedonphysicalswitches)DedicatedNIC(s)forServiceConsoleAtleasttwoforavailabilityNote:easytoconsumemanyphysicalNICsandswitchportsifnotusingVLANtaggingESX服務(wù)器硬件件配置考慮慮要點-存儲應盡可能采采用外置共共享磁盤陣陣列存放虛虛擬機文件件ESX服務(wù)器內(nèi)置置硬盤應有有充分的冗冗余，建議議采用RAID1ESX服務(wù)器自身身對硬盤要要求，安裝裝時的Partition劃分：不建議用安安裝時的自自動硬盤劃劃分方法，，因為/、/var、/home會放再同一一個目錄下下，當/(root)滿了時，ESX服務(wù)器會發(fā)發(fā)生嚴重問問題。建議議：/boot 50到100MB(PrimaryPartition)/ 8.0到18GB(PrimaryPartition)(swap) 2倍的ServiceConsole內(nèi)存,建議固定使使用1.6G/var4GB或更大建議足夠的的ESX服務(wù)器程序序空間大小小為18GB本地端的ISO以及其他文文本文件的的存放空間間要考慮存儲對于虛虛擬機的呈呈現(xiàn)方式7VM層數(shù)據(jù)存儲存儲陣列SCSI控制器虛擬磁盤呈呈現(xiàn)為SCSI控制器SCSI控制器顯示示為BUS或LSILogic磁盤控制器器一個VM可具有1到4個虛擬LSILogic或BusLogicSCSI適配器每個SCSI適配器包含含1到15個虛擬SCSI存儲設(shè)備虛擬磁盤駐駐留在可格格式化為VMFS、NFS或裸磁盤的的數(shù)據(jù)存儲儲中文件系統(tǒng)類類型由底層層物理磁盤盤驅(qū)動器器確確定VMFSNFSFCiSCSINAS卷、數(shù)據(jù)存存儲和LUN卷存儲陣列數(shù)據(jù)存儲8LUN1020GBLUN是一個邏輯輯空間可由存儲陣陣列的整個個空間創(chuàng)建建，也可由由其中的部部分空空間創(chuàng)建LUN映射到ESX后即成為卷卷當卷被格式式化為某種種文件系統(tǒng)統(tǒng)之后即成成為數(shù)據(jù)存存儲不能在同一一個LUN中混用不同同類型的文文件系統(tǒng)每個LUN對應一個VMFS卷虛擬機內(nèi)容容位于數(shù)據(jù)據(jù)存儲中數(shù)據(jù)存儲ESX數(shù)據(jù)存儲類類型：VMware文件系統(tǒng)（VMFS）使用VMFS的RDM網(wǎng)絡(luò)文件系系統(tǒng)（NFS）卷VM內(nèi)容ESX主機數(shù)據(jù)存儲采采用某種文文件系統(tǒng)格格式可以像操作作文件一樣樣操作數(shù)據(jù)據(jù)存儲每個系統(tǒng)具具有256個VMFS數(shù)據(jù)存儲每個系統(tǒng)具具有8個NFS數(shù)據(jù)存儲ISO映像、VM模板和軟盤盤映像9虛擬機內(nèi)容容文件名說明<VM名稱>.vmxVM配置<VM名稱>.vmdk虛擬磁盤<VM名稱>-flat.vmdk預分配的虛擬磁盤文件（包含數(shù)據(jù)）<VM名稱>.vswp交換文件nvram非易失性RAM文件<VM名稱>.vmemVM內(nèi)存<VM名稱>.vmssVM掛起文件<VM名稱>.vmsd快照數(shù)據(jù)<VM名稱>-Snapshot.vmsn快照狀態(tài)文件VMware-0.log、vmware-1.log等日志文件10文件夾/子目錄數(shù)據(jù)存儲類類型數(shù)據(jù)存儲VMFS數(shù)據(jù)存儲VMFS數(shù)據(jù)存儲NFSIP網(wǎng)絡(luò)VM內(nèi)容VM內(nèi)容FC交換機光纖通道SAN磁盤陣列iSCSISAN磁盤陣列NAS磁盤陣列VM1ESX主機1VM2VM1ESX主機2VM2VM內(nèi)容VM3VM312本地SCSIVMDKIP交換機ESX服務(wù)器建議議配置-新購為了盡可能能的發(fā)揮虛虛擬化的作作用，最大大限度的利利用單臺服服務(wù)器的資資源，建議議用于虛擬擬化宿主服服務(wù)器的配配置應達到到或超過如如下標準：：服務(wù)器CPU路數(shù)雙路四路八路CPU（建議主頻2GHz以上）雙路四核四路雙核或四核四路雙核或四核+內(nèi)存16GB+32GB+64GB+千兆網(wǎng)口無外接存儲4+/6+4+/6+4+/6+使用FC存儲4+/6+4+/6+4+/6+使用IP存儲6+/8+6+/8+6+/8+FCHBA口（建議4Gb或8Gb產(chǎn)品）222內(nèi)置硬盤（使用外置磁盤陣列時）222電源雙冗余雙冗余雙冗余從性價比和和可用性考考慮，不建建議在單路路服務(wù)器上上部署虛擬擬化虛擬化宿主主服務(wù)器建建議配置-現(xiàn)有對于目前業(yè)業(yè)內(nèi)用的比比較多的四四路服務(wù)器器，建議議如下：四路單核服服務(wù)器：運運算能力較較弱，虛機機數(shù)量應控控制在10個以內(nèi)，內(nèi)內(nèi)存配置建建議在12GB-16GB；四路雙核服服務(wù)器：運運算能力中中等，虛機機數(shù)量可做做到10-15個左右，內(nèi)內(nèi)存配置建建議在16GB-24GB；四路四核服服務(wù)器：運運算能力強強勁，虛機機數(shù)量可做做到15-30個左右，內(nèi)內(nèi)存配置建建議在24GB-32GB。VC服務(wù)器最佳佳配置建議議處理器：2.0GHz或更高的Intel或AMDx86處理器，VC支持多處理理，可支持持至多2個CPU。內(nèi)存：最低低需求為2GB，假使數(shù)據(jù)據(jù)庫和VC安裝于同一一臺，建議議增加至4GB。磁盤空間：：最小為560MB，建議2GB。網(wǎng)卡：建議議用Gigabit。最低硬件配配置單個2GHzCPU，2GB內(nèi)存，千兆兆網(wǎng)口可支持20個同時連接接，管理50臺物理機，，1000個虛擬機左左右建議配置雙CPU，4GB內(nèi)存，千兆兆網(wǎng)口可支持50個同時連接接，管理200臺物理機，，2000個虛擬機左左右議程應用實施范范圍考慮服務(wù)器采購購考慮虛擬機部署署考慮管理維護考考慮虛機個數(shù)的的規(guī)劃單臺服務(wù)器器所能支持持虛機數(shù)量量的決定因因素：服務(wù)器的硬硬件配置CPU性能多核高主頻頻技術(shù)使得得CPU成為性能瓶瓶頸的可能能性越來越越低內(nèi)存大小做為硬指標標的內(nèi)存，，配置越高高，所能支支持的虛機機數(shù)量越多多網(wǎng)絡(luò)絡(luò)端端口口千兆兆網(wǎng)網(wǎng)環(huán)環(huán)境境已已很很普普遍遍，，網(wǎng)網(wǎng)絡(luò)絡(luò)帶帶寬寬大大多多有有保保證證，，更更多多從從管管理理角角度度來來考考慮慮HBA卡磁盤盤訪訪問問性性能能對對虛虛機機數(shù)數(shù)量量有有一一定定影影響響，，建建議議采采用用4Gb或8GbHBA卡以以減減少少鏈鏈路路影影響響本地地磁磁盤盤內(nèi)置置磁磁盤盤的的可可用用性性及及IO吞吐吐能能力力均均較較弱弱，，不不建建議議在在其其上上存存放放虛虛擬擬機機，，推推薦薦使使用用外外置置高高性性能能磁磁盤盤陣陣列列應用用負負載載大大小小由于于物物理理服服務(wù)務(wù)器器資資源源自自身身的的最最大大限限制制，，應應用用負負載載越越大大，，所所能能同同時時運運行行的的虛虛機機數(shù)數(shù)量量越越少少建議議將將不不同同應應用用訪訪問問特特性性的的應應用用混混合合部部署署在在同同一一物物理理服服務(wù)務(wù)器器上上靈活活運運用用DRS和VMotion技術(shù)術(shù)可可將將物物理理機機與與虛虛機機的的比比率率關(guān)關(guān)系系調(diào)調(diào)到到最最優(yōu)優(yōu)考慮慮到到HA及DRS所要要求求的的資資源源冗冗余余，，所所有有運運行行虛虛機機在在正正常常負負載載下下，，總總體體資資源源使使用用率率不不超超過過三三分分之之二二會會比比較較合合適適經(jīng)驗驗值值：：雙雙路路四四核核10個虛虛機機左左右右，，四四路路四四核核15-30個虛虛機機（（僅僅為為參參考考））虛機機資資源源的的分分配配CPU、內(nèi)內(nèi)存存CPU分配配原原則則：：盡量量使使用用最最少少的的vCPUs，如如果果是是單單線線程程應應用用，，不不支支持持多多線線程程處處理理，，請請不不要要使使用用virtualSMP虛擬擬CPU數(shù)量量不不要要等等于于或或超超過過物物理理CPU核數(shù)數(shù)，，如如雙雙路路雙雙核核服服務(wù)務(wù)器器配配置置的的虛虛機機最最多多使使用用兩兩個個虛虛擬擬CPU當配配置置虛虛擬擬機機的的時時候候須須了了解解ESX服務(wù)器本本身也有有一些overhead。需注意意不要超超過所有有虛擬機機使用率率和所有有vCPU匯總數(shù)目目。觀察”idleloopspin”功能參數(shù)數(shù)，某些些操作系系統(tǒng)當它它們閑置置時，并并不會真真正的釋釋放virtualCPU。確認配置置了單一一處理器器的虛擬擬機為””UPHAL/kernel”，多處理理器的虛虛擬機必必須設(shè)定定為”SMPHAL/kernel”。內(nèi)存分配配原則：：內(nèi)存總量量為在資資源評估估后，計計算虛擬擬機評估估結(jié)果所所需實際際物理內(nèi)內(nèi)存的總總和，其其他由于于應用程程序而產(chǎn)產(chǎn)生的更更多內(nèi)存存需要可可以用ESX的磁盤內(nèi)內(nèi)存來解解決關(guān)鍵應用用可考慮慮固定內(nèi)內(nèi)存的方方法以保保證性能能的穩(wěn)定定性DRSBestPractices:HardwareConfigurationEnsurehostsareCPUcompatibleIntelvsAMDSimilarCPUfamily/SSE3statusEnhancedVMotionCompatibility(EVC)“VMwareVMotionandCPUCompatibility”whitepaperCPUincompatibility=>limitedDRSVMmigrationoptionsLargerHostCPUandmemorysizepreferredforVMplacement(ifallequal)Differencesincacheormemoryarchitecture=>inconsistencyinperformanceDRSBestPractices:ClusterConfigurationHighernumberofhosts=>moreDRSbalancingoptionsRecommendupto32hosts/clusterMayvarywithVCserverconfigurationandVM/hostratioNetworkconfigurationonallhostsVMotionnetwork:Securitypolicies,VMotionnicenabled,GigEnetwork,etcVirtualMachinenetworkpresentonallhostsVMdatastoresharedacrossallhostsVMfloppy/CDconnectedtohostdeviceDRSBestPractices:VMResourceSettingsReservations,Limits,andSharesSharestakeeffectduringresourcecontentionLowlimitscanleadtowastedresourcesHighVMreservationsmaylimitDRSbalancingOverheadmemoryUseresourcepools(RP)forbettermanageabilityVirtualCPU’sandMemorysizeHighmemorysizeandvirtualCPU’s=>fewermigrationopportunitiesConfigureVMsbasedonneedDRSBestPractices:AlgorithmSettingsAggressivenessthresholdModeratethreshold(default)workswellformostcasesAggressivethresholdsrecommendedifHomogenousclustersandVMdemandrelativelyconstantandFewaffinity/anti-affinityrulesUseaffinity/anti-affinityrulesonlywhenneedAffinityrules:closelyinteractingVMsAnti-affinityrules:I/Ointensiveworkloads,availabilityAutomaticDRSmoderecommended(cluster-wide)Manual/Partiallyautomaticmodeforlocation-criticalVMs(perVM)PerVMsettingoverridescluster-widesettingHABestPractices-Setup&NetworkingProperDNS&NetworksettingsareneededforinitialconfigurationAfterconfigurationDNSresolutionsarecachedto/etc/FT_HOSTS(minimizingthedependencyonDNSserveravailabilityduringanactualfailover)DNSoneachhostispreferred(manualeditingof/etc/hostsiserrorprone)RedundancytoESXServiceConsolenetworkingisessential(severaloptions)ChoosetheoptionthatminimizessinglepointsoffailureGateways/isolationaddressesshouldrespondviaICMP(ping)EnablePortFast(orequivalent)onnetworkswitchestoavoidspanningtreerelatedisolationsNetworkmaintenanceactivitiesshouldtakeintoaccountdependenciesontheESXServiceConsolenetwork(s)VMwareHAcanbetemporarilydisabledthroughtheCluster->EditSettingsdialogValidVMnetworklabelnamesrequiredforproperfailoverVirtualmachinesusethemtore-establishnetworkconnectivityuponrestartHANetworkConfigurationAsingleserviceconsolenetworkwithunderlyingredundancyisusuallysufficient:Useateamof2NICsconnectedtodifferentphysicalswitchestoavoidasinglepointoffailureConfigurevNicsinvSwitchforActive/Standbyconfiguration(rollingfailover=““yes””,defaultloadbalancing=routebasedonoriginatingportID)Considerextendingtimeoutvalues&addingmultipleisolationaddresses(*seeappendix)Timeoutsof30-60secondswillslightlyextendrecoverytimes,butwillalsoallowforintermittentnetworkoutagesNetworkredundancybetweentheESXserviceconsolesisessentialforreliabledetectionofhostfailures&isolationconditionsHANetworkConfiguration(Continued)HAwilldetectanduseasecondaryserviceconsolenetworkAddingasecondaryserviceconsoleportgrouptoanexistingVMotionvSwitchavoidshavingtodedicateanadditionalsubnet&NICforthispurposeAlsoneedtospecifyanadditionalisolationaddressfortheclustertoaccountfortheaddedredundancy(*seeappendix)Continueusingtheprimaryserviceconsolenetwork&IPaddressformanagementpurposesBecarefulwithnetworkmaintenancethataffectstheprimaryserviceconsolenetworkandthesecondary/VMotionnetworkBeyondNICteaming,asecondaryserviceconsolenetworkcanbeconfiguredtoprovideredundantheartbeating&isolationdetectionHABestPractices––ResourceManagementLargergroupsofhomogenousserverswillallowhigherlevelsofutilizationacrossanHA/DRSenabledcluster(onaverage)Morenodespercluster(currentmaximumis16)cantoleratemultiplehostfailureswhilestillguaranteeingfailovercapacitiesAdmissioncontrolheuristicsareconservativelyweighted(sothatlargeserverswithmanyVMscanfailovertosmallservers)Todefinethesizingestimatesusedforadmissioncontrol,setreasonablereservationsastheminimumresourcesneededAdmissioncontrolwillexceedfailovercapacitieswhenreservationsarenotset;otherwiseHAwilluselargestreservationspecifiedasthe““slot”size.Ataminimum,setreservationsforafewvirtualmachinesconsidered““average”AdmissioncontrolmaybetooconservativewhenhostandVMsizesvarywidelyPerformyourowncapacityplanningbychoosing““Allowvirtualmachinestobepoweredoneveniftheyviolateavailabilityconstraints”.HAwillstilltrytorestartasmanyvirtualmachinesasitcan.議程應用實施范范圍考慮服務(wù)器采購購考慮虛擬機部署署考慮管理維護考考慮ImpactofVirtualCenterDowntimeComponentImpactExperiencedVirtualMachinesUnaffected,managementrequiresdirectconnectionstoESXServersESXServersUnaffected,managementrequiresdirectconnectionstoESXServersPerformance&MonitoringStatisticsHistoricalrecordswillhavegapsduringoutages,stillavailableviaESXServersVMotionUnavailableVMwareDRSUnavailableVMwareHAAgentsunaffected&providefailoverfunctionality,admissioncontrolunavailable參見《BulletproofVirtualCenter-AGuidetoProtectingVirtualCenter》》VirtualCenterComponentsVirtualCenterServerWebAccessLicenseServerADDomainControllerDNSServerDatabaseServer參見《BulletproofVirtualCenter-AGuidetoProtectingVirtualCenter》》VirtualCenter–RecommendedCollocationCollocationofVirtualCentercomponentsisdesirableformostenvironmentsFocusofthissessionisonprovidingprotectionforthesecomponentsIndustrystandardsolutionsassumedforothercomponentsOneServer,PhysicalorVirtualVirtualCenterServerWebAccessLicenseServerADDomainControllerDNSServerDatabaseServer參見《BulletproofVirtualCenter-AGuidetoProtectingVirtualCenter》VirtualCenterComponents(AdditionalDetails)VirtualCenterService:almoststatelessInformationaboutinventorystoredinthedatabaseSomestatefilesstoredlocallyonVirtualCenterserverWebAccessNostateinformationLicenseServerLicensefilestoredlocally14dayGraceperiodifunavailable參見《BulletproofVirtualCenter-AGuidetoProtectingVirtualCenter》VirtualCenter–LocalConfigurationFilesOneServer,PhysicalorVirtualVirtualCenterServerWebAccessLicenseServerDatabaseServerSSLCertificateLicenseFileConfig.FileUpgradeFiles參見《BulletproofVirtualCenter-AGuidetoProtectingVirtualCenter》Step1forHighAvailability:ProtecttheDatabaseDatabaseoutagewillterminateVirtualCenterserviceAsofVirtualCenter2.0.1Patch2,WindowsServiceManagerwillautomaticallyattempttorestartitevery5minutes,indefinitelyVirtualCenterDatabaseshouldbeindependentlyinstalledandmanagedForlocalavailabilityusethepreferredmechanismforthetypeofdatabasebeingused(VMwareHA,MSCS,Databasespecificmechanisms)Fordisasterrecovery,databaseshouldbereplicatedtoaremotesiteaspartofanoverallDRplan參見《BulletproofVirtualCenter-AGuidetoProtectingVirtualCenter》VCVCBang!FailoverStep2forHighAvailability:ProtectVirtualCenterVMwareHAandMicrosoftClusterServices(MSCS)arethetwomostpopularoptionsOther3rdpartysolutionspossible**Supporteddirectlyby3rdpartyOptiona):VMwareHAVirtualinstancesonlySubjecttosharedstorage/networkconstraintsOnlyrequiressingleOS&applicationinstance;noexplicitreplicationOptionb):MSCSVirtualCenter2.0.2patch2orbeyondPhysicalorvirtualinstancesRequires2identicalOS&applicationinstallations;explicitreplicationoffilesInvolvesadditionalconfigurationefforts&ongoingmaintenance參見《BulletproofVirtualCenter-AGuidetoProtectingVirtualCenter》VirtualCenter:Physicalvs.VirtualPhysicalVirtualBackupsdoneusingtraditionaltoolsBackupspossiblethroughtraditionaltools,VCB,snapshots,cloning,etc.DedicatedserverrequiredDedicatedservernotrequired,resourcescanbesharedwithothervirtualmachinesPerformancelimitedonlybyserverhardwarePerformancefromsharedresources;tuningmaybeneededVirtualCenterwithVMwareHA:Out-of-BandTwoapproachesTwoVirtualCenterinstancesmanageeachother(pictured)BothruninHAclusterEachmanagestheother’sHAclusterSeparateVirtualCenterinstanceisusedtomanage2-nodeHAcluster(notpictured)vpxdvpxdVirtualCenterServermanagestheVMwareHAclusterprovidingitsprotectionWhentheESXhostswithVirtualCenterVMfails,VMisrestartedautomaticallybyHAFailoverfunctionalityprovidedbyHAisindependentfromVirtualCenter(post-configuration)vpxdVirtualCenterwithVMwareHA:In-BandVirtualCenterwithMSCS––PhysicalBestpractice:useMajorityNodeSetquorumwithwitnessshareMaybeusedasgeographicallydispersedclusterfordisasterrecoverysolutionVCDBmaybeusedinanotherclustergrouponthesameclusterRequiresathirdnodeEthernetNetworkvcdbvpxdVirtualCenterwithMSCS––VirtualRequiresuseofquorumdiskclustering:QuorumdiskonthesharedstorageSystemdisksforbothclusteredvirtualmachinesonlocalstorageIncompatiblewithVMotionorVMwareHASANLANvpxdRiseofthePhoenix-DisasterRecoveryVirtualCenterServerDatabaseServerVIServicesVIInventoryDatabaseServerStandardDRSolutionStandbyVirtualCenterServerPrimaryReplicationofstatefilesVirtualCenterDisasterRecoveryOverviewDisasterrecoverysolutionconsistsofthreepieces:VIInventorydataUsethestandardDRsolutionofthedatabasevendorVIServicesColdStandby–Re-installVirtualCenterandrestorelocalconfigurationfilesWarmStandby–Pre-installVirtualCenterandsynchronizelocalconfigurationfiles,butkeep2ndinstancedisconnectedAllotherinfrastructureservices:AD,DNS,etc.Useexistingproduct-specificsolutionsColdStandbyRecoveryProcedureAbletoassignprimary’sIPtostandby?MaintainseparateuptodatecopyoflocalconfigurationfilesInstallfresh

VirtualCenterinstanceInstall

configfiles;connecttostandbyDBDisaster!ESXServerhosts

reconnectautomaticallyRunscripttoreconnectallESXServerhostsDoneyesnoWarmStandbyRecoveryProcedureAbletoassignprimary’sIPtostandby?Keeplocal

configurationfilessynchronizedbetweenprimaryandstandbyDisaster!ESXServerhosts

reconnectautomaticallyRunscripttoreconnectallESXServerhostsDoneyesnoFasterend-to-endrecovery(RTO)AlsoallowsforscriptingandautomationReplicationofconfigurationfilesthroughhost-basedreplicationorbackuptoolsMonitoringIndividualVirtualCenterComponentsEntitySubentityComponentMetricTooltouseVirtualCenterServerVirtualCenterVirtualCenterService-vpxdvpxd.exefoundWMIvpxdisrunningasserviceWMIVirtualCenterCertificatesfilesareexistsWMIsizeWMImodificationdateWMILicenseServerLicenseServerServiceexefileexistsWMIserviceisupandrunningWMILicenseFilesfilesareexistsWMIsizeWMImodificationdateWMIWebServiceWebServiceexefileexistsWMIserviceisupandrunningWMIWebPagehttp://localhostisreachablePerlorvbsscript–HTTPGETcriticalfiles(?)arefoundWMIHostSystem(whereVCisrunningon)SystemCPUload<90%WMISystemdisk<80%FULLWMINetworkOKWMIDatabaseserverODBCConnectionConnectionworksWMIDatabaseintegritySome"select"statementoncriticaltablessqlFileContentsLocationFlexLicensefilesLicensekeysforallserver-basedlicensedfeaturesC:\ProgramFiles\VMware\VMwareLicenseServer\LicensesVirtualCenterConfigurationfileGovernsthebehaviorofVirtualCenterServeranditsinteractionwithESX