H12-921_V1.0 HCIE-Data Center Network (Written) V1.0 Questions and Answers

The virtual machine is associated with the network through a port

Neutron ' s ML2 plug-in components provide Network, Subnet, and Port services

A subnet is an IPv4/IPv6 address segment, and the subnet needs to be mapped to the network

Network is an isolated Layer 2 network, including VLANs and VXLANs

Static routes

BGP

Policy-based routing

OSPF

Loop detection

Single-path detection

Business simulation

Data consistency check

Container images are smaller than virtual machines

Containers start up faster than virtual machines

Containers don ' t need separate kernels

Containers are more isolated than virtual machines

Service connectivity testing

Business simulation

Tenant snapshot rollback

Fault alarms

Carry the source EPG group ID in the reserved field in the VXLAN packet header

In the 802. 1Q header carries the source EPG group ID

Carry the source EPG group ID in the G flag in the header of the VXLAN packet

Carry the source EPG group ID in the VNI field in the VXLAN packet header

PFC deadlock prevention

PFC deadlock detection

AI ECN

DCQCN

XML

HTML

JSON

PROTOBUF

Etcd

Bird

CNI Plugin

VSwitch

You can use it to view the actual physical path of a specific service flow between VMs

You can configure multiple probe packets to be sent at a time

It can be used to detect whether service flows between physical machines are interrupted

Paths can be filtered based on five-tuple information

Accelerate the node

Fat nodes

MPI node

Log in to the node

VXLAN

TRILL

MPLS

L2GRE

DCI-GW requires VXLAN support

DCI-GW does not learn the routes between the service host and the network

Transmission of common VLAN packets between the Fabric-GW and the DCI-GW

The DCI network only passes the underlay route for the DCI-GW

Network service layer

Business presentation layer/collaboration layer

Network Analysis/Control Layer

Storage access layer

Network-wide rollback allows you to roll back the entire network configuration to before the configuration change to quickly restore services

The three-level rollback function supports manual backup and automatic backup, and supports automatic recovery when the network is abnormal

Tenant snapshots are backups and restores of the network at tenant granularity

Service rollback supports single-point service recovery

Headroom cache space actually occupies the Guaranteed cache of the inbound queue

The headroom cache room is fixed at 20M

The headroom cache is used to store packet D. received from the time when the queue sends a PFC backpressure notification packet to before the upstream device stops sending packets

Headroom cache space actually occupies the shared cache of the outbound queue

Label

Instance

Datasheet

Flavor

No more than 50ms

No more than 100ms

No more than 10ms

No more than 1s

The network that carries NSH packets must be a VXLAN network

Through the metadata carried in NSH packets, the packets can be parsed on NSH-Aware SF without encapsulation or decapsulation operations to know the information of the packet tape, which improves the efficiency of packet parsing and improves the forwarding performance

NSH-Unaware SF does not recognize NSH packets and discards them upon receipt

SFC Proxy receives packets from SFR on behalf of SFF and its associated NSH-Unaware representatives, deletes the NSH encapsulation information, and sends packets to through local logical components NSH-Unaware SF

VLAN Hand-off

L3 Underlay

End-to-end VXLAN

Segment VXLAN

BGP and OSPF protocols

Device startup configuration

Number of packets lost on the interface

ARP entries

Cloud-network integration scenario

Container network scenario

Rack rental scenario

Calculate the linkage scenario

Traffic analysis

Verification of intent

Data Collection

Network modeling

SSD

HDD

SCM

Flash

Security group

EIP

Microsegmentation

Business chain

AI ECN

INOF

NPCC

INC

SNMP

NETCONF

WebService

gRPC

When Nova creates a virtual machine, it calls the network interface of Glance to create the virtual machine

Nova supports creating, modifying, and deleting virtual machines

Nova relies on Keystone for certification services

Nova is used to provide virtual computing resource services for the OpenStack cloud platform

iMaster NCE-FabricInsight

Router

Data center switches

SmartNICs

RESTful API

RESFCONF API

NETCONF API

Service Chain API

The controller delivers the orchestrated services to the network device for automatic configuration

When a tenant submits a lease intention to an operator, it must specify the lease form, bandwidth, and value-added services

The tenant obtains the server ' s IP , account number, password, and other information from the carrier and uses the network

Tenants can only use VAS services provided by carriers

Hypervisor

Guest OS

Guest Machine

Host OS

Cinder is generally used for long-term storage of persistent type of data at rest

Cinder provides object storage services

Swift provides block storage services

Swift is generally used to store virtual machine images, data backups, etc

NETCONF

SNMP

OpenFlow

RESTful/RPC

Network-wide rollback

Configure rollback

Tenant rollback

Business rollback

Underlay is logically isolated from overlay, and as long as the routes between tunnel endpoints are reachable, the failure of the underlay network will not directly affect the overlay network

Overlay Routes for the network are generated by routes that are ingested into the underlay network

When implementing the overlay technology in the underlay network, the original underlay network needs to be modified to a certain extent

From the perspective of an overlay network, the underlay network is transparent to devices connected outside of the VXLAN tunnel endpoints

Application layer

Control & Analysis Layer

Server layer

Forwarding layer

In data parallelism, each computing node trains the complete model based on local data

Distributed training can support the training of massive data and complex models

Data parallelism and model parallelism are common distributed training methods

In model parallelism, each compute node will only interact with the trained sub-model and obtain the final model through model aggregation

6

8

2

4

On the target side, the leaf matches the default access control policy of the EPG group members

On the source side, Leaf matches the default access control policy of the EPG group members

On the source side, the leaf matches the default access control policy of the unknown EPG group member

On the target side, the leaf matches the default access control policy of the unknown EPG group member

SFF

SF

sc

SFP

IEEE 802. 1AB

IEEE P802. 1CQ

IEEE 802. 1Qcp

IEEE P802. 1Qcz

Ceilometer

Manila

Trove

Neutron

Static NAT64 requires static configuration of IPv6 and IPv4 address mapping, which does not require refresh or aging

The well-known NAT64 prefix 64: FF9B::/96 already exists by default and does not need to be configured

The custom NAT64 prefix can be of any length, which can be 32, 36, 40, 44, etc

The NAT64 device determines whether the packet performs NAT64 translation based on the IPv6 prefix of the destination address of the IPv6 data packet

Static IPv6 routes

OSPFv3

IBGP4+

EBGP+

Docker

Kubernetes

Swarm

Mesos

1:1

1:N

N:1

N:N

The RP sends an ECN set packet

RP reduces the packet send rate

The RP sends a PFC counterpressure

The RP pauses packet sending

Firewall

WAF

LoadBalance

NIP

Cloud-network integration scenario

Interface with third-party controllers, such as VMwareNSX controllers

underlay, which is configured through a third-party configuration tool such as Ansible

Computing linkage (virtualization) scenarios

The plug-in agent is deployed on every host running the hypervisor and provides a local virtual switch configuration

Compute nodes include: Core Plug-in, Agent, and Service Plug-in

Core Plug-in provides services including Network, Subnet, Port, and Router

The Neutron Server is responsible for receiving the Message Queuing request and invoking the appropriate plug-in to process it

Multi-host sharing is inconvenient

Limited scalability, up to 255 nodes

Direct connection to the CPU is not supported

Can ' t pull it far

Firewall

Firewall Policy

Firewall Rule

FirewalAction

Network service layer

Business presentation/collaboration layer

Network Analysis/Control Layer

Compute the access layer

Hub-Spoke

Full-mesh

Centralized gateways

Distributed gateways

Do

Retry

Rollback

DryRun

In the case of a two-level network, the Spine switch does not have an uplink interface, and all ports are used to connect to the Leaf switch

When designing a network architecture, a bottom-up approach is generally used to determine how many layers of architecture to use based on the size of the network

In two-tier networking, the number of Spine switches is equal to the total number of upstream interfaces of all leaf switches divided by the number of interfaces per Spine switch and rounded up by D

First, the total number of network access interfaces is determined according to the number of computing nodes in the cluster and the number of service interfaces of each node

Active and standby controller clusters

Simulation verification is supported

NSH and microsegmentation

The control supports the deployment of a single controller

TransitRouter

TGW Router

TransitVAS

External gateways

Group members of the same EPG can communicate directly with each other, and there is no restriction on how they cannot communicate with each other

In the cross-device forwarding scenario, the destination EPG group number is written to the VXLAN packet header and forwarded to the remote VTEP

In the local forwarding scenario, the source VM and destination VM are on the same network device, and the EPG packet information is maintained on the network device

GBP is a flow control policy based on EPG packets + protocol number + port number, which specifies the flow control policies within EPG packets and between EPG packets

Autonomous driving

machine learning

artificial intelligence

robot

Computing networks

Storage front-end networking

Storage back-end networking

Business management network

IMaster NCE-Fabric supports backhaul routes delivered to managed third-party VASs to the Fabric network

IMaster NCE-Fabric supports route configuration and VAS service delivery from Huawei VAS to Fabric network

IMaster NCE-Fabric supports the delivery of route configuration from the Fabric network to the VAS side, regardless of whether the VAS is managed by the controller or not

IMaster NCE-Fabric supports business automation orchestration of third-party VAS, as long as the VAS is within the scope of the controller

This solution uses different BDs to achieve inter-tenant isolation to meet the security requirements of multi-tenant isolation

Spine is deployed as a stand-alone L3 device and forms ECMP with Leaf to achieve traffic load balancing

The Leaf switch is deployed as an L3 gateway and needs to enable the ARP Any-Agent feature

Spine-Leaf supports Layer 2 or Layer 3 networking

Type 1 Routing (Intra-ASI-PMSIA-Droute).

5 Class Routing (SourceActiveA-Droute).

6 Class Routing (Share Tree Join route).

7 Class Routing (SourceTreeJoinroute).

Security Group implements basic access control for virtual machine traffic

External Network is a special type of network that enables virtual machines to communicate with external networks

Router is used to connect subnets within a tenant to the same network or between different networks, as well as to connect internal and external networks

The virtual machine can access the external network through the floating IP, but the outside world cannot access the virtual machine through the FLoating IP

Nova

Swift

Ironic

Neutron

Networking

Routing

VPC interworking

Specify the source firewall