华为CloudEngine系列交换机FCoE和DCB技术白皮书

5G-Advanced网络技术演进白皮书（2021）——面向万物智联新时代从产业发展驱动角度看，键，全球的主要经济体均明确要求将5G作为长期产业发展的重要一环。

从业务上5G将要进入千行百业，从技术上5G需要进一步融合DOICT等技术。

因此本白皮书提出需要对5G 网络的后续演进—5G-Advanced进行持续研究, 并充分考虑架构演进及功能增强。

本白皮书首先分析了5G-Advanced的网络演进架构方向，包括云原生、边缘网络和网络即服务，同时阐述了5G-Advanced的技术发展方向包括智慧、融合与使能三个特征。

其中智慧代表网络智能化，包括充分利用机器学习、数字孪生、认知网络与意图网络等关键技术提升网络的智能运维运营能力，打造内生智能网络；融合包括行业网络融合、家庭网络融合、天地一体化网络融合等，实现5G与行业网协同组网、融合发展；使能则包括对5G交互式通信和确定性通信能力的增强，以及网络切片、定位等现有技术的增强，更好赋能行业数智化转型。

，华为，爱立信（中国），上海诺基亚贝尔，中兴，中国信科，三星，亚信，vivo，联想，IPLOOK，紫光展锐，OPPO，腾讯，小米（排名不分先后）1 产业进展概述 (01)1.1 5G产业发展现状 (01)1.2 5G网络演进驱动力 (01)1.2.1 产业发展驱动力 (01)1.2.2 网络技术驱动力 (02)2 5G-Advanced网络演进架构趋势和技术方向 (04)3 5G-Advanced关键技术 (06)3.1 网络智能化 (06)3.1.1 网络智能化关键技术 (06)3.1.2 智能网络应用场景 (08)3.2 行业网融合 (08)3.3 家庭网络融合 (09)3.4 天地一体化网络融合 (10)3.5 交互式通信能力增强 (11)3.6 确定性通信能力增强 (11)3.7 用户面演进 (12)3.8 网络切片增强 (12)3.9 定位测距与感知增强 (13)3.10 组播广播增强 (13)3.11 策略控制增强 (13)4 总结和展望 (14)5G网络的全球商用部署如火如荼。

1接口基础配置关于本章1.1 接口简介通过本小节，您可以了解到设备的接口分类和接口编号规则。

1.2 配置接口基本参数配置接口基本参数，包括接口描述信息、接口流量统计时间间隔功能以及开启或关闭接口。

1.3 维护接口您可以通过清除接口统计信息以方便查询一定时间内接口的流量信息。

1.1 接口简介通过本小节，您可以了解到设备的接口分类和接口编号规则。

接口分类接口是设备与网络中的其它设备交换数据并相互作用的部件，分为管理接口、物理业务接口和逻辑接口三类，其中：l管理接口管理接口主要为用户提供配置管理支持，也就是用户通过此类接口可以登录到设备，并进行配置和管理操作。

管理接口不承担业务传输。

关于管理接口的详细配置，请参见《CloudEngine 7800&6800&5800系列交换机配置指南-基础配置》。

设备支持的管理接口如表1-1所示：表1-1各管理接口介绍l V100R005C00版本下，仅CE6850-48S6Q-HI支持Mini USB接口。

V100R005C10及以后版本，CE6850-48S6Q-HI、CE6850–48T4Q-HI和CE6850U-HI支持Mini USB接口。

l CE6850HI和CE6850U-HI设备上有两个Combo类型的管理接口，每个Combo口包括一个光接口和一个电接口。

光接口和电接口只能同时激活其中一个。

l物理业务接口物理业务接口是真实存在、有器件支持的接口。

物理接口需要承担业务传输。

物理接口有时也被称为端口，为便于描述，在本手册中，统一描述为接口。

设备支持的物理接口如表1-2所示。

表1-2物理接口缺省情况下，设备的以太网接口工作在二层模式，如果需要应用接口的三层功能，可以使用undo portswitch命令将接口转换为三层模式。

l逻辑接口逻辑接口是指能够实现数据交换功能但物理上不存在、需要通过配置建立的接口。

逻辑接口需要承担业务传输。

设备支持的逻辑接口如表1-3所示。

CloudEngine Is the Foundation of the Intent-driven NetworkHuawei CloudEngine Series Switches Technical PresentationContentsClick to add Title 1Click to add Title 2Click to add Title 3CloudEngine Switch OverviewCloudEngine Switch HighlightsCloudEngine Switch Market ProgressSwitches Are the Cornerstone for Transforming Data Centers from Service Centers to Value CentersCloud computingBig DataDistributed storageMetcalfe's Law: The effect of a telecommunications network is proportional to the square of the number of connected users of the system.-Robert Metcalfe who invented Ethernet, founder of 3Com In the future, even if all hardware network devices will disappear, data center switches used as the buses connecting to servers, will always exist. The Ethernet helps release the value of data.AISDN NFVIntent-driven networkUltra-broadbandInfrastructureSimplifiedOpenController and management tool layerControllerAnalyzerSecurityCharacteristics of the Future of DC Switches: Ultra-broadband, Simplified, Intelligent, Secure, and OpenIntelligentSecureManagementControlAnalysisOpenConvergenceEcosystemSpine LeafGatewayBandwidth -> LatencyLayer 2 and Layer 3 -> SecurityManual driving -> Automated drivingWeb page -> Service integrationCloudEngine Series Data Center Switches Portfolio(1)Core SwitchesAccess SwitchesCloudEngine 6881-48S6CQ (New)CloudEngine 6863-48S6CQ (New)CloudEngine 16800 (New)CloudEngine 16816CloudEngine 16808CloudEngine 1680410GE ToR switch25GE ToR switchCloudEngine Series Data Center Switches Portfolio(2)Core SwitchesAccess Switches10GE ToR switchCE6851-48S6Q-HICE6810-48S4Q-LICE6810-32T16S4Q-LIGE ToR switchCE5855-48T4S2Q-EICE5855-24T4S2Q-EIVirtual SwitchesCE1800VCE6855/CE6856-48S6Q-HICE6855/CE6856-48T6Q-HI10GE large-buffer ToR switchCE6870-48S6CQ-EIToR switch with flexible cardsCE8861-4C-EI40GE switchCE7855-32Q-EICE6865-48S8CQ-EI25GE ToR switch100GE switchCE8850-64CQ-EICE12816CE12812CE12808CE12808S CE12804SCE12804CE12800CE12800SCE6870-48T6CQ-EICE6875-48S4CQ-EICE8850-32CQ-EICE6860-48S8CQ-EICE6857-48S6CQ-EICE8860-4C-EICE5880-48T6Q-EICE6880-24S4Q2CQ-EIOrthogonal architectureStrict front-to-back airflow designNon-blocking switchingCounter-rotating fansLine cardNo cabling of the backplaneIncreased bandwidth of the entire systemCE12800Patent No.: CN201110339954.1Independent front-to-back airflowEven heat dissipation, suitable for data centersCell switching and VoQTraffic balancing, improving bandwidth utilizationCounter-rotating and turbo fansHighly efficient heat dissipationLeading energy-conserving designHigh-Quality CloudEngine: High-Quality Architecture Creates a Green and Stable Network•Industry-leading architecture design and high quality:orthogonal SFU design,Clos architecture,cell switching,and Virtual Output Queue (VOQ)mechanism 1/31/31/31/31/31/31/31/31/31/31/31/31/31/31/31/31/31/3ContentsClick to add Title 1Click to add TitleCloudEngine Switch Overview3CloudEngine Switch Market Progress▪Ultra-broadband Cloud Engine ▪Simplified Cloud Engine ▪Intelligent Cloud Engine▪Secure Cloud Engine ▪Open Cloud Engine2CloudEngine Switch HighlightsClos Theory: Cluster Scale Is the Driving Force for Data Center Network Architecture EvolutionSpineLeafCore EdgeSpineLeafCoreEdgeSpineLeafEdgeL2L310GE10GE40GE 40GE 40GEL2BGPL3L3First generation: 3K GE serversSecond generation: 10K GE serversThird generation: 20K 10GE servers⚫The port capacity of cards increases continuously, and CE switches canprovide 36*100GE ports.⚫To avoid HASH polarization, CE switches provide 128 ECMP paths.⚫Network congestion control: CE switches provide large buffer, and split a single flow into multiple ones to load balance them.Difficulties in the non-blocking Clos architecture: The convergence ratio and packet loss ratio cannot be compromised.⚫Data center network architecture: A fat-tree topology is used and the capacity of the root node determines the server cluster scale.⚫Evolution direction: Add network layers, and increase the quantity and capacity of spine or code nodes.⚫Network congestion control: Increase the buffer and optimize load balancing.CE switches' buffer is 80 times higher than the industry average, implementing zero packet loss for microburst traffic. The switches' performance is 1032 Tbit/s. CE switches can connect to over 50,000 servers with no blocking.Larger Interface Rate: The Rise of 25GE Interfaces Balance the Cost and Efficiency10M1980100M: IEEE802.3u199519981000M: IEEE802.3ab/z20021G: IEEE802.3ae/ak201040G/100G: IEEE802.3ba2008DCB/PB (IETF TRILL)2013400G2009FCoE201625G: 802.byDevelopment of Ethernet:DPDKIn the past two years, why are 25GE interfaces used?➢The 25GE interface can better match the SerDes rate: 1.25 Gbit/s -> 3.125 Gbit/s -> 6.25 Gbit/s -> 10.3125 Gbit/s -> 25 Gbit/s -> 56 Gbit/s➢Compared with the 40GE NIC, the 25GE NIC has higher use efficiency of the PCIe channel. (40G+40G)/8G*16= 62.5%; 25G*2/(8G*8) = 78%➢Lower cabling costs for 25GE interfaces: The SFP28 module is used. Because only single-channel connections are used, the SFP28 module is compatible with LC optical fibers in the 10GE era, without cabling.➢The bandwidth between NICs has exceeded 10 Gbit/s: As technologies such as RDMA, SR-IOV, and DPDK develop, the bandwidth between NICs has exceeded 10 Gbit/s.APP RDMA NICCoprocess or/FPGAAPPRDMA NICFast CNPCNPPPVIQ: eliminates packet lossinside chipsVIQ12Dynamic ECN3Dynamic ECNFast CNPPacket lossTraditionalVIQVIQ enables the outbound interface to send backpressure signals to the inbound interface, achieving zero packet loss.PFCSRECNNormal CNPPhysical queueThreshold (Port-Buffer)SwitchServer Q0Q1ECN waterline ECN waterlineRR ServerServerServer.. .Dynamic ECN uses dynamic collection and dynamic threshold adjustment to realize low latency and high throughput.Fast CNP provides fast congestion feedback to improve networkconvergence performance by 30%.AI Fabric: Intelligent Lossless Data Center Network Solution Provides Low latency and Zero Packet LossFast CNPUltra-broadband Cloud Engine Simplified Cloud Engine Intelligent Cloud Engine Secure CloudEngine Open Cloud Engine➢Independent forwarding, control, anddetection, and 3-channel cluster design ➢Four dedicated GE interfaces are used ascluster control channels.➢ A maximum of 3.2 Tbit/s cluster bandwidthis supported.Unique three-channel separated cluster, control plane coupling…Control signalingchannelData forwardingchannel Dual-active detection (DAD) channel➢The control plane runs independently and synchronizes asmall amount of information about interface status entries.➢Devices in the DFS group can be upgraded independently,without interrupting services.➢When the peer-link is faulty or the M-LAG master devicefails twice, the M-LAG backup device can still work properly.Independent control plane, protocol-level coupling…Peer-linkDAD channelDevice Virtualization: Easy-to-Manage, High-Performing, Highly Reliable Virtual SystemsC luster S witch S ystem (CSS)M ulti-Chassis LAG (M-LAG)➢The control plane runs independently anddoes not have synchronization information.➢Two switches are configured with the samegateway IP address and MAC address.➢Two links of the server NIC are configured tosend broadcast packets simultaneously.Independent control planewithout couplingM-LAG Lite➢Provide a maximum of 1:16virtualization capabilities in port and port group mode.On-demand VS allocation, improving resource utilization➢Exclusive CPU, memory, and MAC/VLAN/FIB entriesExclusive resources in VSs and highest specificationsFault isolation between VSs, improving securityVS (V irtual S ystem )Ultra-broadband Cloud Engine Simplified Cloud Engine Intelligent Cloud Engine Secure CloudEngine Open Cloud EngineLayer 2 Boundary Extension: Build a Large-Scale Network Resource Pool Based on BGP EVPNBGP EVPN acts as the VXLAN control plane to provide the following functions:➢Triggers automatic VXLAN tunnel setup between VTEPs to avoid the need to manually configure full-mesh tunnels.➢Advertises host routes and MAC address table, prevents unknown traffic flooding, and optimizes packet forwarding.➢Implements Layer 2 interconnection between data centers in different networking.Layer 2 large-scale horizontal expansion in the data centerand extension to the remote DCVXLAN BGP EVPNVTEP VTEP VTEP VTEP RRRRBGP EVPNVTEP VTEPProtocol vitality: open interconnection and interworkingbetween devices from different vendorsNetwork Automation: Interconnection with Third-Party Management Tools, Controllers, Virtualization Management Platforms, and Cloud PlatformsScenario 1: traditional network management➢Interconnection with a third-party management tool : CE switches can interconnect with a third-party management tool such as Ansible to implement automatic network configuration.Scenario 2: network and computing association➢Interconnection with a virtualization management platform : CE switches are connected to the Agile Controller-DCN, and the Agile Controller-DCN is associated with the third-party computing management platform.Scenario 3: third-party management on the overlay➢Interconnection with a third-party controller : The CE switch functions as the VXLAN Layer 2 VTEP and is managed by the NSX.Scenario 4: cloud-network integration➢Interconnection with a cloud platform through the Agile Controller-DCN : CE switches are connected to the AgileController-DCN, and the Agile Controller-DCN connects to the third-party cloud platform.SpineLeafGateway10GE40GEDCScenario 1Interconnection with a managementtoolISP2Scenario 2Interconnection with avirtualization managementplatformScenario 3Interconnectionwith a third-party controller Scenario 4Interconnection with a cloud platformSimplified Deployment: IPv4 and IPv6, Unicast and Overlay Multicast, and Rollout of Full-stack Services Within MinutesServer leaf IPv6IPv4VTEP Server leaf SpineVTEPServer leaf VM VM VMOVSBorder-LeafBMVTEP VM VM VMOVSServer leaf VTEPBMIPv4 extranetIPv4Service-centered IPv6 evolution mode➢2018 Q3: virtualization ➢2019 Q1: cloud-network cooperationServer leafIPv6IPv4VTEP Server leaf SpineVTEPServer leaf VM VM VMOVSBorder-LeafBMVTEP VM VM VMOVSServer leaf VTEPIPv6 extranetIPv4 extranetIPv4IPv6 extranetReplicate IPv4 O&MexperiencesServer leafVTEP Server leaf SpineVTEPServer leaf VM VM VMOVSBM VTEP VM VM VMOVSServer leaf VTEPIPv4 extranetIPv4 extranetOverlay multicastSave bandwidthServer leafVTEP Server leaf SpineVTEPServer leaf VM VM VMOVSBM VTEP VM VM VMOVSServer leaf VTEPIPv4 extranetIPv4 extranetDual-stackMulticastIngress replicationIGMP/PIM-SM➢2018 Q3: commercial chip ready ➢2019 Q1: controller mappingCollectorAnalyzerCPU Forwarding Chip CollectorAnalyzerCPUForwarding ChipCollectorAnalyzerCPUForwarding ChipNP•SNMP or NETCONF uses the query/response mechanism, minute-level reporting, and XML or text encoding, which is inefficient.•NetStream uses the flow sampling mechanism and requires CPU participation, which has low performance and is inaccurate.•gRPC uses the subscription/reporting mechanism, subsecond-level reporting , protobuffer coding, and HTTP transmission, which has a high efficiency .•ERSPAN+ adds ingress and egress ports or timestamps of original flows to calculate the flow path and delay .•INT supports in-line path or quality detection.SNMP NetconfNetstreamERSPANSNMP NetconfNetstreamERSPAN+INTgRPCSNMP NetconfFlow tableProtobuf over UDPgRPCERSPAN+Netstream1:1•Protobuf over UDP is used to encode andtransmit forwarding plane information, which is efficient and does not affect CPU performance .•Small NP intelligent analysis algorithm is used to perform in-depth analysis of abnormal flows to learn in-depth information such as the latency, jitter, packet loss ratio, and packet loss location .Historical CapabilitiesCurrent CapabilitiesFuture EvolutionTelemetry Capability: Transformation of the Data Collection Mode Is the Basis of Big Data O&MUltra-broadband Cloud Engine Simplified Cloud Engine Intelligent Cloud Engine Secure CloudEngine Open Cloud EngineMicroburst Detection Capability: Millisecond-level Buffer Monitoring and Subscription Collection, Which Are Visible and ClearData CenterService exceptionNormal networkTraditional NMSArtifactFreezeSubsecond-levelcollectiongRPC subscription . . .Content feedbackOne request formultiple tasks•Visible: subsecond-level subscription data collection510us50%100%The buffer is full and packet loss may occur.2-ms buffer monitoring•Clear: high-precision data monitoring5-minute pollingperiodSNMP request and response. . .Multiple requests for a single task510s50%100%Normal buffer detectionMicrosecond-level buffer monitoring•The collection period is too long, which may ignore network details.•The detection interval is too long, so device details may be incomplete.Forwarding chipCPUFPGACE8860Monitoring queueNote: The CE8860 supports this function.VM 1VM 2VM 3 1.1.1.1 1.1.1.2 1.1.1.3VM 4VM 5VM 6 2.2.2.1 2.2.2.2 2.2.2.3As Is: subnet-based isolation To Be:VM-level isolationFine-grained DefenseDefining applications based on VM names and discrete IP addresses, with finer granularity and wider dimensionsFlexible DeploymentDefining services based on application groups and decoupling from subnets to achieve flexible deployment Distributed SecurityTraffic of access switches is filtered nearby and east-west isolation is implemented without using firewalls.Use Microsegmentation to Achieve Fine-grained Isolation and Service SecurityWebAppAFW IDS LB NATVASResource poolSimplified deploymentThe SDN controller definesservice chains through drag-and-drop operations.Efficient forwardingProvide traffic diversion for one time, simple configuration, service traffic forwarding, and secure monitoring.Flexible orchestrationDecouple the VAS function from Fabric, providing flexible orchestration.Switch SwitchSwitchAgile CloudEngine: Supporting NSH Service Chains, Providing Easier VAS OrchestrationACOpticalfiber/transmissiondevice/Layer 2 transparenttransmissionMacSec at the Link Layer: IP Layer 3 Features such as Encryption Are Introduced to the MAC Link LayerSwitch ASwitch BScenario➢In scenarios that require high data confidentiality, such asgovernment, military, and finance scenarios, interconnection is required between data centers or between different modules of data centers across buildings.➢The CE6875 uplink port (100GE), and CEL16CQFD (16*100GE) and CEL08CFFG1 (8*200GE) cards of the CE12800 can be used.Definition➢Media Access Control Security (MACsec) ensures securecommunication within LANs in compliance with IEEE 802.1AE and 802.1X. It provides identity authentication, data encryption, integrity check, and replay protection to protect Ethernet frames and prevent devices from processing attack packets.NetworkingNetworkingOriginal packetData encryption protectionMACsec packetData integrity protectionStandardsChinaIntegration InnovationEcological cooperationGermany MoscowMulti-vendor pre-integration verificationMulti-layer open ecosystemOpen Ecosystem: Huawei Joins Hands with 20+ Industry Chain Partners to Perform System IntegrationOpen ecosystem: fast integration andsimplified managementSystem integration: 10+ OpenLabs in the globeManufacturerNSXAnsible. . .Rapid response to service requirementsHardware BFDMicrosegmentationNSH modeIPv6 over VXLANCPUForwarding chipIntra-card CPU chip Quad-core CPU:▪Protocol packet processing▪FIB entry delivery ▪. . .Co-processor▪Hardware BFD ▪High-performance sFlow ▪. . .Forwarding chipAdjustable processes New service processes Adjustable entry resourcesEnhanced serviceprocessesVRPnetconfCLILinux ContainergRPCopenflowSSHpuppetFuncEditnetconfSNMPLinux and driverFragmentation and reassemblyOpen architecture, Flexible Business Innovation•Higher interface rate: 25GE interfaces and larger buffer cope with traffic surge in N:1 scenarios.•Flowlet&DLB: One flow is load balanced among multiple links.•AI Fabric intelligent lossless data center network solution: low latency and zero packet lossUltra-broadband: higher interface rate, more even load balancing, larger buffer, and lower latency•Telemetry capability•Microburst detection •Edge analysis capability•Microsegmentation used to isolate east-west traffic on switches (east-west traffic is isolated on firewalls originally)•SFC used to divert traffic from the control plane to the data plane•MACsec hardware encryption, providing high security and reliability•Open API•Interconnection with third-party management tools: Ansible•Interconnection with third-party management tools or controllers: VMaare NSX•Multiple virtualization technologies: CSS, M-LAG, M-LAG Lite, and VS •VXLAN + BGP EVPN: intra-DC and inter-DC virtualization•SDN controller: deployment in drag-and-drop mode, IPv4 and IPv6, rollout of unicast and multicast full-stack services in minutesIntelligent: enabling service agilitySimplified: automatic deployment of full-stack services and service rollout within minutesOpen: easy integration and timelyresponse to servicesSecure: best quality in the industry and pioneering energy-saving technology CloudEngine High-Performance Cloud SwitchesContents1Click to add Title 2CloudEngine Switch OverviewCloudEngine Switch Highlights Click to add Title 3CloudEngine Switch Market ProgressChina's No.1 and One of World's Top 3 DCN Vendors2014Source: IHS “2015 Infonetics Data Center and Enterprise SDN Vendor Leadership Analysis ”20132012•Data center network vendor with the fastest growth•First release of InterOP impressing the world •Industry-leading ultra-high performance2015•Huawei was the only Chinese vendor in the global SDN leadership list.•Largest market share in China in Q2•Huawei was the global data center network vendor with the fastest growth.•Annual growth rate up to 137%Global SDN AuthoritativeReport of Leading Vendors2016•The market share ranks No. 1 in Chi na and the third largest in the world .•SDN capability won the Best of ShowNet Award at Tokyo Interop.2017•Huawei was positioned as challenger in Gartner's Magic Quadrant for Data Center Networking.2018•Huawei has been positioned as a leader in data center hardware platforms for SDN.•The AI Fabric won the Best of Show Gold Award.In 2013, the CloudEngine 12800 won the Best of Show Award at Interop, which is the highest exhibition in the IT industry. Huawei is the first Chinese provider that wins the position.Highly Recognized PerformanceAward of Excellent Product Trusted byCIOAward of the Most Competitive Product Awards and CertificationsPreferred Brand of Cloud Computing and Network SolutionAward of Annual Excellent TechnologyChina SDN SDN Best Practice AwardAward of Excellent Product in Big DataIn 2016, the CE8860 and CE6851 won the Best of Show Award at Interop.InterOP AwardsHuawei's AI FabricIntelligent Lossless Data Center Network Solution Takes Home Interop Tokyo Best of Show AwardCloudEngineSeries Switches Serve 7800+ Global Customers⚫The market share is No.1in China and No.3in theworld.⚫The global market share growth rate is No. 1 for fourconsecutive years .⚫Over 32,000 CE12800 switches have been soldaround the world, serving 7800+customers in 120+countries.DC SDN SDN hardware platform leader⚫2018 Approaching the Leaders Quadrant ⚫2017 ChallengerGartner Peer InsightsCustomers’ Choice for Data Center NetworkingCopyright©2018 Huawei Technologies Co., Ltd.All Rights Reserved.The information in this document may contain predictivestatements including, without limitation, statements regarding the future financial and operating results, future productportfolio, new technology, etc. There are a number of factors that could cause actual results and developments to differ materially from those expressed or implied in the predictive statements. Therefore, such information is provided for reference purpose only and constitutes neither an offer nor an acceptance. Huawei may change the information at any time without notice.把数字世界带入每个人、每个家庭、每个组织，构建万物互联的智能世界。

HUAWEI TECHNOLOGIES CO.,LTD.华为云存储服务技术白皮书huawei2012/6/26目录1 华为云存储服务介绍 (5)1.1 简介 (5)1.2 服务定位 (5)2 华为云存储应用场景 (6)2.1 网盘服务 (6)2.2 互联网海量数据存储 (7)2.3 数据备份/容灾 (8)2.4 数据归档 (9)2.5 BigData/HPC (9)3 华为云存储服务概念及功能介绍 (11)3.1 云存储基本概念 (11)3.1.1 对象（Objects） (11)3.1.2 存储空间 (12)3.1.3 REST接口 (12)3.2 云存储功能 (12)3.3 访问权限控制 (13)3.4 线下大数据量导入/导出服务 (14)3.5 对象的版本化管理 (15)3.6 可定制的日志服务 (15)4 华为云存储服务特点 (16)4.1 易用性 (16)4.2 高扩展性 (16)4.3 高可靠性 (17)4.3.1 数据可靠性 (17)4.3.2 数据校验 (18)4.3.3 数据自动修复 (18)4.4 高可用性 (19)4.4.1 网络级可靠性设计............................................... 错误!未定义书签。

4.4.2 节点可靠性设计 (19)4.4.3 进程可靠性设计 (19)4.5 高性能 (19)4.5.1 并发访问性能 (20)4.5.2 吞吐能力 (20)4.6 安全性 (20)4.6.1 软件控制安全 (20)4.6.2 数据分块存储（暂未实现） (21)4.7 易管理性 (21)4.8 按需计费 (21)5 华为云存储服务结构简介 (22)5.1 华为云存储服务框架简介 (22)5.2 UDS结构简介 (23)5.2.1 UDS系统结构描述 .............................................. 错误!未定义书签。