Huawei H19-402_V1.0 - HCSP-Presales-Data Center Network Planning and Design V1.0

In Huawei data center networking, especially within Intelligent Lossless Data Center Network design, incast traffic is a well-known challenge in distributed systems such as AI computing clusters and distributed storage architectures .

In an incast scenario , a large number of servers simultaneously send traffic to a single destination (e.g., parameter server in AI training or storage node in distributed storage). This sudden aggregation of traffic can quickly overflow switch buffers , leading to packet loss, increased latency, and throughput degradation .

Huawei identifies incast as a primary cause of network congestion in modern data centers. To mitigate this, CloudFabric solutions implement advanced congestion control technologies such as:

PFC (Priority Flow Control) to prevent packet loss

ECN (Explicit Congestion Notification) for early congestion signaling

AI-driven traffic scheduling and load balancing

These mechanisms help build a lossless Ethernet fabric , which is essential for RDMA-based AI workloads and high-performance storage systems.

Therefore, the statement is TRUE , as incast traffic is indeed prone to causing congestion in such environments.

In Huawei routing principles, route selection follows a strict hierarchy. When multiple routing protocols advertise routes to the same destination, the system first compares the route preference (priority) , not the cost. The route with the lowest preference value (highest priority) is selected. Only within the same routing protocol are cost/metric values compared to determine the best route.

Option A is correct because only one optimal route is installed in the routing table at a time, based on protocol preference. Option B is also correct: direct routes are automatically generated for networks connected to local interfaces. Option C is valid as Huawei devices support route import (redistribution) between routing protocols, enabling flexible network design.

Option D is incorrect because it reverses the decision logic. It incorrectly states that cost is compared before priority, which contradicts Huawei’s routing selection process.

Therefore, the false statement is D .

Huawei data center solutions emphasize the transition toward all-flash data centers as a key enabler of enterprise digital transformation. Traditional HDD-based storage systems cannot meet the performance requirements of modern applications such as AI, big data analytics, cloud computing, and real-time services .

SSD (Solid-State Drive) technology provides significant advantages over HDDs, including:

Ultra-low latency (microseconds vs milliseconds)

High IOPS (Input/Output Operations Per Second)

Better reliability and durability (no mechanical parts)

Lower power consumption and higher density

As a result, SSDs have become the mainstream storage medium in modern data centers, especially in Huawei’s all-flash storage solutions (e.g., OceanStor Dorado) .

All-flash data centers enable:

Faster application response times

Improved user experience

Efficient support for AI and distributed workloads

Simplified storage architecture

Huawei promotes integrating all-flash storage with intelligent lossless networks (RoCE-based) to achieve end-to-end high-performance infrastructure.

Therefore, the statement is TRUE .

In Huawei CloudFabric architecture, a PoD (Point of Delivery) or site is defined as an independent network fault domain under the management of the SDN controller (such as iMaster NCE-Fabric). This design is fundamental for large-scale data center deployments, ensuring clear fault isolation, simplified management, and scalable expansion.

Each PoD typically consists of one or more spine-leaf fabric architectures , which provide high-bandwidth, low-latency, and horizontally scalable connectivity. Within a PoD, all devices operate under a unified control and policy framework delivered by the controller. If a failure occurs within a PoD, it is contained and does not affect other PoDs, thereby improving overall network reliability.

Huawei recommends this modular design approach to support multi-data center deployments, disaster recovery, and service expansion. Multiple PoDs can be interconnected to form a larger cloud fabric, while still maintaining independent fault domains.

Therefore, the statement accurately reflects Huawei CloudFabric design principles and is TRUE.

In Huawei Cloud Stack (HCS) architecture, ManageOne serves as the unified cloud management platform , responsible for end-to-end orchestration, service provisioning, and lifecycle management across compute, storage, and network resources. It provides a centralized portal for administrators and tenants, enabling automated deployment and O & M of cloud services.

FusionSphere is Huawei’s virtualization platform (similar to OpenStack + hypervisor integration) and operates at the resource layer, not as the top-level management system. Hypervisor and Virtual Machine Monitor (VMM) are components of the virtualization layer that manage virtual machines on physical servers, but they lack cloud-wide orchestration capabilities.

From a Huawei CloudFabric perspective, ManageOne integrates with controllers like iMaster NCE-Fabric to automate network provisioning, enabling seamless underlay and overlay configuration. It supports multi-data-center management, self-service portals, and policy-driven automation, which are essential for modern cloud data center operations.

Therefore, ManageOne is the correct answer as the cloud management platform in HCS.

The key difference between M-LAG and stacking in Huawei data center design lies in their control plane architecture and fault domains .

Option A is false because M-LAG does NOT have a centralized control plane . Instead, each device in an M-LAG pair maintains its own independent control plane , while synchronizing state information (such as MAC and ARP tables) over the peer-link. This distributed control design improves reliability and avoids a single point of failure.

Option B is correct : M-LAG involves two independent switches , both of which must be managed (though automation tools like iMaster NCE-Fabric simplify this).

Option C is correct : M-LAG provides fault domain isolation . A failure on one device does not impact the control plane of the other, enhancing network resilience.

Option D is correct : In contrast, stacking uses a single control plane , so upgrades can be more complex and potentially disruptive, affecting the entire stack.

Huawei best practices favor M-LAG over stacking in modern data centers due to its distributed control, higher reliability, and better fault isolation .

Therefore, the correct answer is A .

In Huawei CloudFabric architecture, server leaf nodes and service leaf nodes have distinct roles, even though both can function as NVEs (Network Virtualization Edge devices) in a VXLAN network.

A server leaf node primarily provides access for compute resources , such as physical servers and virtual machines. It acts as a common NVE, handling VXLAN encapsulation/decapsulation for tenant traffic and enabling communication within the overlay network.

However, firewalls and load balancers (LBs) are classified as value-added service (VAS) devices and are typically connected to service leaf nodes , not standard server leaf nodes. Service leaf nodes are specifically designed to handle service insertion, traffic steering, and policy-based forwarding required for these devices.

Although in some simplified deployments roles may be combined, Huawei standard design clearly separates:

Server leaf → compute access

Service leaf → VAS device access (FW, LB, etc.)

Therefore, the statement is FALSE , as server leaf nodes do not typically provide access for firewalls and load balancers in standard CloudFabric architecture.

In Huawei VXLAN architecture, all listed statements accurately reflect core design concepts used in CloudFabric deployments.

A is correct: a Virtual Interface (VBDIF/VirtualIF) is a Layer 3 interface associated with a Bridge Domain (BD) , enabling inter-subnet routing (distributed gateway function).

B is correct: Huawei VXLAN commonly uses a 1:1 mapping between VNI and BD , where the BD acts as the Layer 2 forwarding domain. This mapping ensures clear separation of tenant networks and simplifies forwarding logic.

C is correct: VXLAN encapsulation uses VTEP (VXLAN Tunnel Endpoint) IP addresses . The source IP is the local VTEP, and the destination IP is the remote VTEP. Together, they logically represent the VXLAN tunnel used for transporting overlay traffic across the IP underlay.

D is also correct: a VAP (Virtual Access Point) represents the service access interface. On Huawei CE switches, VAPs are typically implemented using VLAN-based access , meaning VLANs act as the entry point for VXLAN services.

These concepts are fundamental in Huawei EVPN-VXLAN fabrics, enabling scalable, multi-tenant, and automated data center networking.

In Huawei CloudFabric VXLAN design, distributed gateway architecture is a key enhancement over traditional centralized gateway models, especially for large-scale data centers.

Option A is correct because distributed gateways (typically deployed on leaf nodes) only maintain local ARP/MAC entries for directly connected hosts. This significantly reduces ARP table pressure compared to centralized gateways, where a single device must learn all entries. This improves scalability and performance , which is critical in multi-tenant environments.

Option C is also correct as Huawei leaf switches can simultaneously act as Layer 2 VXLAN gateways (bridging) and Layer 3 VXLAN gateways (routing) . This enables distributed inter-subnet routing directly at the access layer, reducing latency and improving east-west traffic efficiency.

Option B is incorrect because it describes a limitation of centralized gateways , where traffic must traverse a central node, leading to suboptimal paths. Distributed gateways eliminate this issue.

Option D is also incorrect as it again describes centralized gateway constraints, not distributed ones.

Thus, distributed VXLAN gateways provide better scalability, optimal forwarding paths, and flexible deployment , making A and C correct .