Huawei HCSP-Presales-Data Center Network Planning and Design (H19-402) Certification Sample Questions

01. A regional bank's draft fabric connects each leaf to two spines over routed point-to-point uplinks, with ECMP across them. To save cost, each leaf's uplink to one spine runs at 100 Gbit/s and its uplink to the other spine at 40 Gbit/s, and both are given the same routing cost. The customer wants traffic spread so that no uplink congests while another still has headroom.
What should the presales engineer change in the underlay?
a) Give each leaf equal-speed uplinks to every spine
b) Keep the mixed speeds and add a second 40 Gbit/s uplink toward the slower spine
c) Raise the routing cost of the 40 Gbit/s uplink to balance the two paths
d) Bundle both uplinks into one Eth-Trunk spanning the two spines

02. A government agency's design team is drafting the IP address plan for a new VXLAN spine-leaf fabric. They want underlay addressing kept clearly separate from tenant address spaces, management addressing and the overlay, and ask which addresses belong in the underlay routing plan.
Which two items should the underlay plan include?
(Choose two.)
a) The Layer 2 VNIs, advertised as routes so spines can forward per tenant
b) The tenant server subnets, advertised in the underlay routing protocol
c) A loopback address per device, used as router ID and VTEP source
d) The anycast gateway address of each tenant subnet, one per leaf
e) Point-to-point subnets for the leaf-to-spine interconnect links

03. Several servers in one VLAN of an e-commerce data center intermittently lose connectivity to their gateway. A packet capture shows gratuitous ARP replies from an unknown host claiming the gateway's IP address, and the servers' ARP caches briefly map the gateway IP to that host's MAC address.
Which attack does this evidence indicate?
a) MAC address table flooding
b) ARP spoofing
c) DHCP server spoofing
d) TCP SYN flooding

04. A logistics firm's procurement lead is reviewing a CloudFabric proposal. She asks whether the solution will also deliver the virtual machines, containers and bare-metal servers that host the firm's warehouse applications, so that the firm can cancel a separate compute tender.
How should the presales engineer answer?
a) No; those workloads sit in the server layer, and CloudFabric does not provide server layer resources
b) Yes; the server layer is supplied by CloudFabric together with the switches and the controller
c) Partly; CloudFabric provides the containers, while VMs and physical servers come from the tender
d) Yes; iMaster NCE-Fabric creates the VMs and containers, so only the physical servers need a separate tender

05. A university runs its CloudFabric data center as one management domain, with one set of iMaster NCE-Fabric managing its first POD. It is adding a second POD on the same campus, and services must be provisioned across both PODs. A consultant insists that a multi-data center controller (MDC) is needed for this.
Which component should provision services across both PODs?
a) A new MDC layered above both PODs
b) One iMaster NCE-Fabric set per POD
c) The cloud platform directly
d) The existing iMaster NCE-Fabric

06. A bank runs two DCs active-active, each with a stateful perimeter firewall that does not share session state with the other. After cutover, some client sessions fail: requests enter through DC1, but the server replies leave through the firewall in DC2.
What is the root cause?
a) Missing EVPN Type 3 routes for the client subnets
b) Too few ECMP paths from the servers toward DC2
c) Asymmetric routing through stateful firewalls
d) PFC pause frames discarding replies at the firewall

07. A payment processor's converged storage network carries RoCEv2 storage traffic and ordinary TCP management traffic on the same links. The design must stop the sender of storage traffic when a downstream buffer fills, without halting management traffic on that link.
Which mechanism provides this behavior?
a) Ethernet Pause on the whole interface
b) PFC on the storage priority
c) ECN marking on the storage queue
d) Strict-priority scheduling for the storage queue

08. An AI training cluster spans several racks connected by a routed spine-leaf fabric, with each rack in its own IP subnet. The customer wants GPU servers in different racks to exchange RDMA traffic over this Ethernet fabric without stretching a Layer 2 domain between racks.
Which transport fits this requirement?
a) RoCEv2, because it runs over UDP/IP
b) iWARP over a stretched VLAN, because RDMA needs one broadcast domain
c) RoCEv1, because it uses its own Ethertype inside each rack's VLAN
d) Native InfiniBand, because it can be carried in each rack's Ethernet subnet

09. A government cloud requires that service provisioning for its fabric resumes quickly even if the entire site hosting the network controller is lost, for example in a fire. Its current proposal runs the fabric's SDN controller as a local cluster in that site only.
Which change best meets the requirement?
a) Add more nodes to the local controller cluster inside the same site and room
b) Rely on M-LAG between leaf pairs to protect the controller servers' network links
c) Use the cloud platform in the second site to configure the switches directly
d) Add a disaster recovery deployment of the controller in a second site

10. A securities firm's fabric uses IBGP EVPN, with its two spines configured as route reflectors for all leaves. A design reviewer worries that concentrating route exchange on the spines makes the control plane fragile, and asks what the reflector design actually gains over having the leaves peer directly.
Which two statements correctly assess this design?
(Choose two.)
a) The spines reflect EVPN routes, so leaves need not peer in a full mesh
b) Two reflectors keep routes flowing if one spine fails
c) The spines must become VTEPs to reflect routes for each tenant
d) The spines must run EBGP to the leaves to reflect EVPN routes
e) The spines replicate BUM traffic for the leaves they reflect routes to