Exchange郵件系統(tǒng)部署Nutanix超融合的最佳實(shí)踐

1、Exchange郵件系統(tǒng)部署Nutanix超融合的最佳實(shí)踐Challenges deploying Exchange NO Virtualization2Dependency on the hardware platformeg: Device Drivers / No mobilityPhysical hardware must be oversized to cater for growthAdditional silo/s of resources are created & need to be managedDelayed time to market for new servers

2、 due to procurementGenerally inefficient use of compute and/or storage resourcesVirtual EnvironmentShared StorageExchange StorageExchange ComputeChallenges deploying Exchange NO Virtualization3Increased demands in the datacentereg: Rackspace , Power, Cooling , Network & Storage commsLimited/No abili

3、ty to scale compute resources without downtimeIncreased complexity/cost forHardware Maintenance / RepairBusiness Continuity & Disaster RecoveryPatching and general BAU tasksChallenges deploying Exchange NO Virtualization4Increased demands in the datacentereg: Rackspace , Power, Cooling , Network & S

4、torage commsLimited/No ability to scale compute resources without downtimeIncreased complexity/cost forHardware Maintenance / RepairBusiness Continuity & Disaster RecoveryPatching and general BAU tasksMORE OPEX!Misconceptions about Virtualizing Exchange5Microsoft do not support Exchange on VMwareRed

5、uced or Poor performanceMust use Direct Attached StorageMust have dedicated spindles / storage for ExchangeExchange does not work on NFSvSphere features such as HA/DRS & vMotion are not supportedExchange DAGs* are not supported for VMsNo availability benefits when using DAGsPoor performance on Thin

6、Provisioned VMDKsSupport options for MS products running in VMs/support/policies/ms_support_statement.htmlSupport policy for MS software running in non-Microsoft hardware VMs/kb/897615* Database Availability GroupsAdvantages of Virtualizing Exchange 6No dependency on Hardware typeEg: Avoids purchasi

7、ng/sizing for forecasted 3-5 year requirementsReduced CAPEX - Purchase only HW required for initial deploymentAbility to save on CAPEX and avoid over sizing and wasted investmentReduced time to market when deploying/expanding Exchange env.New Exchange servers can be stood up in hours not weeks/month

8、sIncreased ROI from Compute/Storage/NetworkHigher utilization levels for servers and network with minimal wastageAdvantages of Virtualizing Exchange7Ability to quickly/easily scale without downtimeEg: ESXi hosts can be added to the cluster without downtimeLower Datacenter & OPEX costsEg: Less rack s

9、pace, power , cooling , equipment , maintenance contracts etcUse vMotion / Live Migration to achieve zero downtime HW maintenanceReduced complexity for Business Continuity & Disaster RecoveryAbility to scale out Exchange server roles without additional HWBenefits of using Snapshots / Cloning & other

10、 Hypervisor featuresAdvantages of Virtualizing Exchange on Nutanix8Further Reduced CAPEX Just 3 node initially then 1 node at a timeStart with just 3 nodes, and scale one node at a time forever! Ability to scale compute & storage linearly where required with the option of compute or storage heavy no

11、des while avoiding Increased Ability to scale quickly/easily without downtimeEg: Nutanix nodes can be added without downtime, increasing the entire clusters performanceNo delay in having capacity available due to having to expand RAID groups etcAdvantages of Virtualizing Exchange on Nutanix9Further

12、Reduced time to market when deployingexpanding the Infrastructure to support ExchangeNew Nutanix clusters can be built in 30 minsNew Nutanix nodes can be added in 95% of the time during BAUWrite I/O is always local w/ 2nd copy (if required) distributed across the clusterMBX SRV 1MBX SRV 2MBX SRV 3VM

13、VMVMStep 1 - From the VM across the hypervisor to the Nutanix Controller1Advantages of Virtualizing Exchange on Nutanix Data Locality13Data Locality feature ensures low latency/high performanceVirtual machines data is hosted on the node where the VM is running Results in higher performance & Reduced

14、 dependency on the IP network Read I/O will be served locally 95% of the time during BAUWrite I/O is always local w/ 2nd copy (if required) distributed across the clusterMBX SRV 1MBX SRV 2MBX SRV 3VMVMVM122Step 2 The Nutanix Controller retrieves data from Extent Cache (DRAM) OR from local SSD or SAT

15、A (only for “Cold Data”)Advantages of Virtualizing Exchange on Nutanix Data Locality14Data Locality feature ensures low latency/high performanceVirtual machines data is hosted on the node where the VM is running Results in higher performance & Reduced dependency on the IP network Read I/O will be se

16、rved locally 95% of the time during BAUWrite I/O is always local w/ 2nd copy (if required) distributed across the clusterMBX SRV 1MBX SRV 2MBX SRV 3VMVMVM1223Step 3 The Nutanix Controller send the request back to the VM over the ESXi hypervisor with NO impact to the rest of the MS Exchange environme

17、nt.Advantages of Virtualizing Exchange on Nutanix Solves “Noisy Neighbor”15Traditional SAN/NAS suffers from the “Noisy Neighbor” issueVMs on different hosts, can impact the performance of other workloads on the shared storage20K IOPS20K IOPS20K IOPS20K IOPS20K IOPS100K IOPS PerformanceExample of a t

18、raditional SAN/NAS SolutionAdvantages of Virtualizing Exchange on Nutanix16Natively Eliminates the “Noisy Neighbor” issueTraditional Exchange deployments on SAN use dedicated RAID packs for Exchange, but even with dedicated spindles, all IO goes via a limited number of storage controllers resulting

19、in contention20K IOPS20K IOPS20K IOPS20K IOPS20K IOPS100K IOPS PerformanceWhat happens when 1 VM demands more IOPS?Advantages of Virtualizing Exchange on Nutanix 17Natively Eliminates the “Noisy Neighbor” issueTraditional Exchange deployments on SAN use dedicated RAID packs for Exchange, but even wi

20、th dedicated spindles, all IO goes via a limited number of storage controllers resulting in contention10K IOPS10K IOPS60K IOPS10K IOPS10K IOPS100K IOPS PerformanceThe entire environment is impacted!Advantages of Virtualizing Exchange on Nutanix18Natively Eliminates the “Noisy Neighbor” issueTraditio

21、nal Exchange deployments on SAN use dedicated RAID packs for Exchange, but even with dedicated spindles, all IO goes via a limited number of storage controllers resulting in contention10K IOPS10K IOPS60K IOPS10K IOPS10K IOPS100K IOPS PerformanceHow does Nutanix solve this problem?Advantages of Virtu

22、alizing Exchange on Nutanix19Natively Eliminates the “Noisy Neighbor” issueTraditional Exchange deployments on SAN use dedicated RAID packs for Exchange, but even with dedicated spindles, all IO goes via a limited number of storage controllers resulting in contention10K IOPS10K IOPS60K IOPS10K IOPS1

23、0K IOPS100K IOPS PerformanceHow does Nutanix solve this problem?Advantages of Virtualizing Exchange on Nutanix20Nutanix uses a “Scale out Shared nothing” ArchitectureVMs on one host are serviced by a dedicated Virtual Storage Controller (Nutanix CVM)VMs on different nodes do not impact each otherThi

24、s is a native feature - No need for hypervisor specific features (ie: Storage I/O Control)VMVMVMVMVM20K IOPS20K IOPS20K IOPS20K IOPS20K IOPSAdvantages of Virtualizing Exchange on Nutanix21Nutanix uses a “Scale out Shared nothing” ArchitectureVMs on one host are serviced by a dedicated Virtual Storag

25、e Controller (Nutanix CVM)VMs on different nodes do not impact each otherThis is a native feature - No need for hypervisor specific features (ie: Storage I/O Control)Now what happens when a VM demands more IOPS?VMVMVMVMVMVMVMVMVMVM20K IOPS20K IOPS20K IOPS20K IOPS20K IOPSAdvantages of Virtualizing Ex

26、change on Nutanix22Nutanix uses a “Scale out Shared nothing” ArchitectureVMs on one host are serviced by a dedicated Virtual Storage Controller (Nutanix CVM)VMs on different nodes do not impact each otherThis is a native feature - No need for hypervisor specific features (ie: Storage I/O Control)IOP

27、S can burst on one Nutanix node with minimalOr no impact to other nodes & Virtual machinesVMVMVMVMVM20K IOPS20K IOPS30K IOPS20K IOPS20K IOPSAdvantages of Virtualizing Exchange on Nutanix23Multi-Hypervisor SupportvSphere , Hyper-V and KVM all supportedMix a match within the same Nutanix cluster for o

28、ptimal ROI / Utilization Storage PoolNutanix features remain consistent across hypervisorsNo need to change your Infrastructure to suit a specific hypervisorNo Vendor lock in!Advantages of Virtualizing Exchange on Nutanix 2 Data Protection OptionsTwo data protection options for DAG and Non DAG deplo

29、ymentsReplication Factor (RF) 1 for DAGs with 3 or more copies = 25% capacity savings over RF2 2 DAG copiesReplication Factor (RF) 2 for non DAG deployments, or DAGs with 2 copiesBoth RFs can be used concurrently on the same underlying storage pool25Ability to mix workloads without risk to SLAsMailb

30、ox servers can reside on different nodes with low I/O VMs to ensure optimal performance No requirement for Exchange to have dedicated SSDs/HDDs/RAID packs etcExchange shares a single Storage Pool for maximum capacity utilizationAbility to share capacity with workloads such as VDI without compromisin

31、g performanceAdvantages of Virtualizing Exchange on Nutanix Mixed WorkloadsAdvantages of Virtualizing Exchange on Nutanix 26Ability to remove or replace compute & storage without downtimeRemove old compute & storage nodes without downtimeEasier to manage than DAG on Physical hardware with DAS storag

32、eLess BAU staff requiredLess time “Keeping the lights on” & more time for “Innovation”No need for specialized or dedicated solution for ExchangeRun Exchange on new or existing Nutanix cluster/s with mixed workloads!Single Pane of glass Management for Compute & StorageLess BAU overheadsLess training

33、requiredEasier TroubleshootingOnly need to look after the Hypervisor & Nutanix not half a dozen windows openDifficult/Costly to scale Capacity/Performance linearlyTraditional shared Storage Performance is DIVIDED between Compute nodesAdding a Compute node, further DIVIDES storage performance resourc

34、esPerformance/Scaling Challenges traditional SAN50K IOPS50K IOPS100K IOPS Performance1000 Mailboxes1000 MailboxesPerformance Great! Difficult/Costly to scale Capacity/Performance linearlyTraditional shared Storage Performance is DIVIDED between Compute nodesAdding a Compute node, further DIVIDES sto

35、rage performance resourcesPerformance/Scaling Challenges traditional SAN33K IOPS33K IOPS33K IOPS100K IOPS Performance1000 Mailboxes1000 Mailboxes1000 MailboxesPerformance per user reduced Difficult/Costly to scale Capacity/Performance linearlyTraditional shared Storage Performance is DIVIDED between

36、 Compute nodesAdding a Compute node, further DIVIDES storage performance resources25K IOPS25K IOPS25K IOPS25K IOPS100K IOPS Performance1000 Mailboxes1000 Mailboxes1000 Mailboxes1000 MailboxesPerformance per user 50% of original solution Performance/Scaling Challenges traditional SANDifficult/Costly

37、to scale Capacity/Performance linearlyTraditional shared Storage Performance is DIVIDED between Compute nodesAdding a Compute node, further DIVIDES storage performance resources20K IOPS20K IOPS20K IOPS20K IOPS20K IOPS100K IOPS Performance1000 Mailboxes1000 Mailboxes1000 Mailboxes1000 Mailboxes1000 M

38、ailboxesPerformance per user Totally degraded! Performance/Scaling Challenges traditional SANDifficult/Costly to scale Capacity/Performance linearlyTraditional shared Storage Performance is DIVIDED between Compute nodesAdding a Compute node, further DIVIDES storage performance resources20K IOPS20K I

39、OPS20K IOPS20K IOPS20K IOPS100K IOPS PerformanceWhat about adding disk shelves?1000 Mailboxes1000 Mailboxes1000 Mailboxes1000 Mailboxes1000 MailboxesPerformance/Scaling Challenges traditional SANDifficult/Costly to scale Capacity/Performance linearlyTraditional shared Storage Performance is DIVIDED

40、between Compute nodesAdding a Compute node, further DIVIDES storage performance resources20K IOPS20K IOPS20K IOPS20K IOPS20K IOPS100K IOPS PerformanceWhat about adding disk shelves?1000 Mailboxes1000 Mailboxes1000 Mailboxes1000 Mailboxes1000 MailboxesPerformance per GB of Storage REDUCES!Performance

41、/Scaling Challenges traditional SANVMVMVM1000 Mailboxes1000 Mailboxes1000 Mailboxes25K IOPS25K IOPS25K IOPS75K IOPS Performance/Scaling with NutanixVMVMVM1000 Mailboxes1000 Mailboxes1000 MailboxesVM1000 Mailboxes25K IOPS100K IOPS (In 2RU!)25K IOPS25K IOPS25K IOPSPerformance/Scaling with NutanixVMVMV

42、M1000 Mailboxes1000 Mailboxes1000 MailboxesVMVM1000 Mailboxes1000 Mailboxes25K IOPS25K IOPS125K IOPS 25K IOPS25K IOPS25K IOPSPerformance/Scaling with NutanixVMVMVM1000 Mailboxes1000 Mailboxes1000 MailboxesVMVMVM1000 Mailboxes1000 Mailboxes1000 Mailboxes25K IOPS25K IOPS25K IOPS150K IOPS w/ Linear Sca

43、lability25K IOPS25K IOPS25K IOPSPerformance/Scaling with NutanixScale one Node at a time and INCREASEStorage ControllersRead Cache (Extent Cache DRAM)Read/Write Cache (Flash Storage)Storage Capacity (SATA)Compute Capacity (CPU/RAM)VMVMVM1000 MailboxesPerformance Remains Great at any scale! 1000 Mail

44、boxes1000 MailboxesPerformance/Scaling with NutanixVMVMVMVMVMFailure ScenariosABCABC5 Node N+1 Cluster functioning normally (Dedicated Exchange or mixed workload cluster)Hyper-Converged Availability for ExchangeVMVMVMVMVMABCFailure Scenarios5 Node N+1 Cluster functioning normally (Dedicated Exchange

45、 or mixed workload cluster)One node fails The Hypervisor restarts Exchange VM on a surviving host and/or Exchange DAG failoverExchange VMs functions normally accesses data locally/remotely and/or DAG passive copy becomes activeHyper-Converged Availability for ExchangeVMVMVMVMVMABCABCFailure Scenario

46、s5 Node N+1 Cluster functioning normally (Dedicated Exchange or mixed workload cluster)One node fails vSphere HA restarts Exchange VM on a surviving host and/or Exchange DAG failoverExchange VMs functions normally accesses data locally/remotely and/or DAG passive copy becomes activeNutanix detects s

47、ome data does not have two copies (for RF2 deployments) & repairs clusterHyper-Converged Availability for ExchangeVMVMVMVMVMABCABCFailure Scenarios5 Node N+1 Cluster functioning normally (Dedicated Exchange or mixed workload cluster)One node fails vSphere HA restarts Exchange VM on a surviving host

48、and/or Exchange DAG failoverExchange VMs functions normally accesses data locally/remotely and/or DAG passive copy becomes activeNutanix detects some data does not have two copies (for RF2 deployments) & repairs clusterVM data is localized over time as data is read over the 10G LAN Hyper-Converged A

49、vailability for ExchangeVMVMVMVMVMABCABCFailure Scenarios5 Node N+1 Cluster functioning normally (Dedicated Exchange or mixed workload cluster)One node fails vSphere HA restarts Exchange VM on a surviving host and/or Exchange DAG failoverExchange VMs functions normally accesses data locally/remotely

50、 and/or DAG passive copy becomes activeNutanix detects some data does not have two copies (for RF2 deployments) & repairs clusterVM data is localized over time as data is read over the 10G LAN Cluster is fully functional and fully repaired and able to tolerate subsequent failuresOnly 20% of the cont

51、rollers are lost (vs 50% for legacy storage) Exchange Environment fully online!Hyper-Converged Availability for ExchangeVMVMVMVMVMABCA 2nd host fails The hypervisor restarts Exchange VM on a surviving host and/or Exchange DAG failoverExchange VMs functions normally accesses data locally/remotely as

52、requiredNutanix detects some data does not have two copies & repairs clusterVM data is localized over time as data is read over the 10G LAN Cluster is fully functional and fully repairedCluster is fully functional and fully repaired and able to tolerate subsequent failuresOnly 40% of the controllers

53、 are lost (vs 100% for legacy storage) Exchange Environment fully online!ABCFailure ScenariosHyper-Converged Availability for ExchangeCapacity PlanExchange Server Role Requirements CalculatorInput the exchange environments configuration. For example, the number of mailbox, mailbox size，DAG parameter

54、s, the number of copies. This tool will automatically calculate the number of servers and storage capacity and performance requirements.InputOutputComputing resource requirementsStorage resource requirementsSizerTransform resource requirements into configurationExchange on AHVCapital Expenditure Sav

55、ings Unlike other hypervisors and management solutions, Acropolis and AHV come free with every Nutanix node, thereby eliminating the so-called virtualization tax. AHV management components do not require purchase or installation of Tier 1 database platformsall required management components are buil

56、t into the distributed platform and scale automatically as clusters expand. AHV provides Exchange deployments with all the required virtualization features below at no cost: High availability and live migration. Hardware abstraction. Performance monitoring. Simplicity Run Nutanix with AHV in a fully optimal configuration, from out of th