版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
1、August 2013FlexPod Select with HadoopTechnical Customer PresentationThe Big Data Analytics Gap2Existing SkillsApplicationsDataGrowthSource: IDCLost Value“Big data” refers to datasets whose size is beyond the ability of current tools to capture, store, manage, and analyzeChallenges with Hadoop in Ent
2、erprise3Operations ImplementationRequires three copies of data, larger footprint, and more storageLimited flexibility; storage and servers tied together affects scalability Low cluster efficiency, higher network congestionNameNode is a single point of failureSlow recovery from disk drive failureExpe
3、nsive process to replace failed disks onlineMost common Hadoop support issue is disk drive failureAvailabilityNeed to keep up with fast-paced patches, projects of open source platformNeed to decide on distribution of Hadoop Skills are not commonIntegration with existing IT infrastructure can be diff
4、icultTuning expertise needed to make Hadoop perform optimallyBig Data Opportunities4Fraud detection and preventionAnti-money-laundering Risk managementSupply-chain optimization Defect tracking Root-cause analysisRFID correlationLaw enforcementCounterterrorismDrug developmentPatient recordsEvidence-b
5、ased medicineFinancial ServicesManufacturingHealthcareGovernmentBig Data: Problems Being Solved TodayFinancial InstitutionsTick data analysisTrend analysisRisk assessmentCross-domain correlationsHealth CareDrug effaceDisease pattern recognition and progressionFraud detectionClaim automationGovernmen
6、t (FBI,CIA,DOD,DOE,HLS,IRS,SEC)Internet threat detectionPattern recognitionImage analysis Fraud and waste mitigationConsumer and Commercial SpaceSocial media and product sentiment correlationConsumer analysisAdvertising affinity correlationTelemetry and quality analysis5Changing the Way Business Is
7、ConductedValue Proposition6Some Problems Require an Enterprise-Class Hadoop SolutionEnterprise Class HadoopPackaged ready-to-deploy modular Hadoop cluster The data has intrinsic $ value Usable capacity must expand faster than compute Higher storage performanceReal human consequences if the system fa
8、ils (threats, treatments, financial losses)System has to allow for asymmetric growthWhite Box HadoopValues associated with early adopters of Hadoop Social media space Contributors to Apache Strong bias to JBODSkeptical of all vendorsEnterprise-Class HadoopPackaged ready-to-deploy modular Compute- an
9、d memory-intensive Hadoop cluster Compute-intensive applicationsTick data analysisExtremely tight service-level expectations Severe financial consequences if the analytic run is lateEnterprise Class HadoopBounded compute algorithm and memory-intensive Hadoop cluster Compute-intensive applicationsAdd
10、itional CPUs do not improve run timeExtremely tight service-level expectations Severe financial consequences if the analytic run is lateNeed for deeper storage per DataNodeCompute PowerStorage Capacity7FlexPod FamilyFlexPod Converged Infrastructure Prevalidated, Flexible, Unified Platform8Cisco Unif
11、ied Computing SystemProgrammable infrastructureAbstraction of bare metal server elementsOpen API for automationOrchestration through industry-standard toolsUnified managementPolicy-based automation and service profilesSelf-integrating componentsUnified fabricVirtualization awarenessScalability witho
12、ut complexityHigh-performance I/ONetApp Storage Systems: FAS with Clustered Data ONTAP ; E-Series with Dedicated ApplicationsUnified storageVirtualize storage into pools Nondisruptive data motionIntelligent data managementBuilt-in storage efficiencyIntegrated data protectionHigh performance, dedicat
13、ed workloadsDensity, scale for big dataThroughput for video, media, HPCFlexPod Converged Infrastructure Family50 PBStorage PoolNetwork PoolCompute Pool10,000scoresAppAppAppAppAppAppEnterprise/Service ProviderMSB/Branch OfficeDedicatedBasic Cisco UCS ServersAppDistinct ArchitecturesDistinct Architect
14、uresFlexPod ExpressFlexPod Data CenterFlexPod SelectTestedFull ValidationTested or full validationFAS2220, 2240 (currently 7-Mode) FAS and Data ONTAP (clustered and 7-Mode)Any: E-Series, FAS, FlashRayCisco UCS C-Series, Nexus 3K Cisco UCS, Nexus, UCS FI, UCS Manager Cisco UCS, Nexus, Catalyst, MDS6,
15、000+ commercial resellers800 FlexPod; 44+ premium partnersSpecialized and vertical resellersTwo fixed pod sizesFlexible pod sizesReference architecture and/or designs1GbE fabric10GbE fabricDirect, FC, and 10GbE Cisco UCS Director, VMware, and MicrosoftFlexPod validated management and ecosystemApplic
16、ation-based managementFile SystemMassively scalable shared virtual data center infrastructureBig data analytics, scientific, HPCFor smaller, less-dynamic requirements and VAR velocityDirect or FabricAppApp9FlexPod Data Center Versus FlexPod Select 10FlexPod Data Center (Classic)Architecture includes
17、 Cisco UCS B- or C-Series servers managed by UCS Manager and Fabric InterconnectNetApp FAS storage connected to N5K or N7K access layer (which is connected to UCS Fabric Interconnect)Storage connectivity through fabricData protocol is FC/FCoE or Ethernet-based iSCSI/NFS/CIFSTargeted for data center
18、enterprise applications such as Microsoft apps, VDI, SAP, Oracle, and so onFlexPod Select Architecture includes Cisco UCS C-Series servers only, managed by UCS Manager and Fabric InterconnectPrimarily NetApp E-Series storage for datastore E-series connected to Cisco UCS C-Series servers directly (SA
19、S attached)Storage connectivity through direct-attach to serverData protocol is SASTargeted for Hadoop and big data workloads such as Cloudera and HortonworksNetApp FAS is used for NameNode store (not for datastore)Extending FlexPod to Big Data11Cisco UCS Rack-Mount ServersCisco UCS Blade ServersExt
20、endable to multiple data center implementations for disaster recovery and business continuityCisco UCS ManagerDeploy, Manage, MonitorCisco Tidal Enterprise SchedulerHadoop ConnectorsBig DataEcosystemNetApp FASEnterprise ApplicationsAvailabilityBackupNetApp SnapshotFlexPod Select with Cloudera Enterp
21、rise-grade Hadoop management applicationEnd-to-end administration in a single toolProven at scale to work in production environmentsBuilt-in intelligence, best practices, real-time queriesCisco Validated DesignMasterExpansion Management Software & Technical Support (Included)CDHCMCloudera ManagerCSC
22、loudera SupportOSSApache Hadoop & Open Source Software12FlexPod Select with Hortonworks Deploy seamlessly across multiple operating systems, clouds, virtual platformsThe only100% open source and complete distributionEnterprise grade, proven and tested at scaleEcosystem endorsed for interoperabilityM
23、asterExpansion13Solution Architecture14Solution ArchitectureData ONTAPFAS222015FlexPod Select with Hadoop ArchitectureE5460 storage5 disk trays60 NL-SAS drives6Gb/s SAS (direct attached to servers)FAS2220-2 Data ONTAP 8.1.2 7-ModeCopy of HDFS metadataNFS served to NameNode and secondary NameNodeFAS
24、222016Front View of Master and Expansion RacksMasterExpansionE5460 storage arraysData and Task NodesInfrastructure,NameNode,secondary NameNode,JobTrackerCisco UCS 6296UP (FI)Cisco Nexus 2232PP (FEX)FAS222017Back View of Master and Expansion RacksMasterExpansionE5460 Storage ArraysData and Task Nodes
25、Infrastructure,NameNode,secondary NameNode,JobTrackerCisco UCS 6296UP (FI)Cisco Nexus 2232PP (FEX)FAS222018E5460 Storage Layout for HDFSTwo 7-disk RAID groups with two LUNs per nodeDedicated set of disks per DataNodeShared-nothing architectureSpare disks shared globallyRAID 519FAS2220 Design for Had
26、oop NameNodeMetadata protection through NFSHighly reliable, proven NetApp NFSPrimary and secondary NameNodes have NFS access to FSImage dataQuick turnaround to rebuild NameNode in case of failurePrimary NameNodeSecondary NameNodeNFS 10GbENFS 10GbEFAS222020Architecture OverviewFlexPod Select with Had
27、oop21What Is FlexPod Select with Hadoop?The Cisco Platform for NetApp Open Solution for Hadoop is an implementation of the NetApp Open Solution reference design using the Cisco Platform Architecture for Big Data.Like the NetApp Open Solution for Hadoop v1.0, the Hadoop Rack is based on building bloc
28、ks of Hadoop data and compute servers using NetApp E5460 storage arrays instead of internal local hard-disk drives.The basic building block is four servers for each E5460 storage array.22FlexPod Select with Hadoop ConfigurationMaster rackInfrastructure node; Hadoop NameNode, secondary NameNode, and
29、JobTracker node3 E5460 storage arrays supporting 12 Hadoop data and compute nodes12 Hadoop DataNode and TaskNode serversLocal DNS/NTP serverFAS2240-2 storage array for Hadoop NameNode metadata backupIn-rack 10GbE and 1GbE network infrastructureExpansion rackFour E5460 storage arrays supporting 16 Ha
30、doop DataNodes and TaskNodes16 Hadoop data and compute node serversIn-rack 10GbE and 1GbE network infrastructureNote: Each Hadoop cluster can have only one master rack, but it may have one or more expansion racks.23Why Have a Local DNS Server?BenefitsEliminates the need to modify other enterprise or
31、 data center DNS serversIs easily integrated into enterprise DNS servicesHelps isolate the Hadoop cluster from other enterprise network namespacesServices provided for the Hadoop clusterBoth forward and reverse host name and IP address resolution for the Hadoop clusterA platform for a running NTP ti
32、me service so that all Hadoop cluster resources have a good synchronized time sourceA local copy of NetApp SANtricity Storage Manager for administration and monitoring of the E5460 storage arraysA platform for automated storage provisioning scripts during the racks initial build Possible future serv
33、er servicesService to rebuild failed servers with correct OS version and required packages, such as Kickstart and PuppetHadoop management software24FlexPod Select with Hadoop Networking25Component10GbE ConnectionsGbE ConnectionsE5460 arraysNoneConnection to each of two controllers in the arrayFAS222
34、0Four 10GbE portsConnection to baseboard management Controller and remote LAN module management portNodesTwo 10GbE portsInfrastructure serverTwo 10GbE portsIn-Cabinet Network AdvantagesProvides better performance with 10GbE than traditional GbE Hadoop networksDedicates 10GbE bandwidth to where it is
35、 requiredOffers ease of Hadoop network administration:Most Hadoop network administration is performed on the 10GbE network in the master rackComponent (management) 1GbE administration is performed on the 1GbE switch in the local rack of the affected componentsIsolates Hadoop network traffic from oth
36、er enterprise network trafficMinimizes network cabling to external network infrastructureFacilitates rack network securityUses VLANs to separate administration traffic from Hadoop cluster traffic26Storage System Components27Two Main Storage ComponentsE-Series E5460 storage system FAS2220 storage systemFAS2220,16x1 TBE5460, 4U 60 x3TB/4TBDense Innovative Design for Hadoop4U, sixty 3.5 drive high-density SAS enclosure (up to 240TB via 4TB drives)Five12-drive horizontal drawers = 60 drivesTwo E5400 controllers and SAS HICsOnline front serviceabl
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 九鼎奖金制度
- 仓库领料人员奖惩制度范本
- 医疗废物检查奖惩制度
- 人社所工作奖惩制度范本
- 养老院服务考核奖惩制度
- 班级班规小学奖惩制度
- 专职教师奖惩制度规定
- 办公室每周总结奖惩制度
- 检测公司绩效奖惩制度
- 深化落实安全奖惩制度
- 安全环境职业健康法律法规文件清单(2025年12月版)
- 中华财险2026秋季校园招聘备考题库及答案详解1套
- 《房屋构造(第2版)》教学课件01初识建筑构造
- 2025小红书医美行业精准获客与营销增长白皮书
- 急诊护理安全管理课件
- 国际金融(江西财经大学)学习通测试及答案
- 2025广西投资集团有限公司招聘4人笔试历年备考题库附带答案详解试卷3套
- 2026年湖南生物机电职业技术学院单招职业倾向性考试必刷测试卷必考题
- 2025年驻马店辅警招聘考试真题附答案详解(完整版)
- 化学试题卷答案【中国第一高中】【湖北卷】湖北省2025年华中师大一附中2025年高考学科核心素养卷暨考前测试卷(最后一卷)(5.31-6.1)
- 祖国不会忘记二声部合唱简谱
评论
0/150
提交评论