NCP-AIN Questions and Answers

The correct answer is PXN (Peer eXchange Network) .

From the NVIDIA NCCL Documentation :

"PXN enables communication between GPUs connected via NVLink and NICs by treating the GPUs as a distributed switch. This architecture improves bandwidth utilization by enabling any GPU to communicate with the NIC via the shortest path available, even if it's not directly connected to the NIC."

This enhances GPU-to-NIC and NIC-to-GPU transfers, leveraging the NVLink topology. It significantly boosts performance in multi-GPU setups where not every GPU is directly connected to the NIC.

Other options:

Adaptive Routing is a fabric-level feature for dynamic path rerouting.

Graph Search Optimization is used internally for topology modeling in NCCL.

SHARP v2 is a switch-based collective acceleration method, unrelated to PXN.

The Network Traffic Map in NVIDIA's Unified Fabric Manager (UFM) provides a visual representation of the network topology and traffic flows, which is particularly useful for identifying congestion points, verifying network configurations, and optimizing workload distribution.

However, when troubleshooting a single node's hardware failure, the Network Traffic Map is less effective, as it focuses on network-level issues rather than individual hardware components.​

In multi-tenant AI cloud environments, ensuring that each tenant's workloads are isolated and secure is paramount. The NVIDIA Spectrum-X platform addresses this need through its Traffic Isolation capabilities. This feature ensures that network resources are partitioned effectively, preventing unauthorized access and interference between tenants. By implementing Traffic Isolation, the provider can maintain strict boundaries between different tenant environments, ensuring both security and performance consistency.​

Reference Extracts from NVIDIA Documentation:

"Spectrum-X enhances multi-tenancy with performance isolation to ensure tenants' AI workloads perform optimally and consistently."​

"Spectrum-X utilizes the programmable congestion control function on the BlueField-3 hardware platform to accurately assess the congestion condition of the traffic path by using in-band telemetry information... to achieve the goal of performance isolation to ensure that each tenant gets the best expected performance in the cloud and is not negatively affected by congestion of other tenants."​

The CloudAI Benchmark is designed to evaluate and establish performance baselines in AI-optimized networking environments like NVIDIA Spectrum-X. It assesses various performance metrics, including throughput and latency, ensuring that the network meets the demands of AI workloads. This benchmarking is essential for validating the benefits of Spectrum-X and for ongoing performance monitoring.​

To enable SHARP (Scalable Hierarchical Aggregation and Reduction Protocol) aggregation using the NCCL-SHARP plugin , the following two environment variables are required:

NCCL_COLLNET_ENABLE=1

Enables NCCL’s support for CollNet (Collective Network) operations, including SHARP.

NCCL_SHARP_AUTOINIT=1

Automatically initializes the SHARP plugin when available, activating SHARP-based collectives.

From the NVIDIA NCCL User Guide – SHARP Plugin Section :

"NCCL_COLLNET_ENABLE must be set to enable collective network acceleration features."

"NCCL_SHARP_AUTOINIT enables automatic SHARP plugin integration at NCCL runtime."

Incorrect Options:

B. NCCL_ALGO=CollNet – This variable controls the algorithm used for collectives but does not enable SHARP.

C. NCCLSPECTRUM_ENABLE – This is not a documented NCCL variable.

The NVIDIA Spectrum-X networking platform is an Ethernet-based solution optimized for AI workloads, combining Spectrum-4 switches, BlueField-3 SuperNICs, and advanced software to deliver high performance and low latency. Gathering telemetry data is critical for optimizing Spectrum-X networks, as it provides visibility into network performance, congestion, and potential issues. The question asks for the tool used to collect telemetry data in a Spectrum-X network.

According to NVIDIA’s official documentation, NVIDIA NetQ is the primary tool for gathering telemetry data in Ethernet-based networks, including those running on Spectrum-X platforms with Cumulus Linux or SONiC. NetQ is a network operations toolset that provides real-time monitoring, telemetry collection, and analytics for network health, enabling administrators to optimize performance, troubleshoot issues, and validate configurations. It collects detailed telemetry data such as link status, packet drops, latency, and congestion metrics, which are essential for Spectrum-X optimization.

Exact Extract from NVIDIA Documentation:

“NVIDIA NetQ is a highly scalable network operations tool that provides telemetry-based monitoring and analytics for Ethernet networks, including NVIDIA Spectrum-X platforms. NetQ collects real-time telemetry data from switches and hosts, offering insights into network performance, congestion, and connectivity. It supports Cumulus Linux and SONiC environments, making it ideal for optimizing Spectrum-X networks by providing visibility into key metrics like latency, throughput, and packet loss.”

— NVIDIA NetQ User Guide

This extract confirms that option C, NetQ, is the correct tool for gathering telemetry data in a Spectrum-X network. NetQ’s integration with Spectrum-X switches and its ability to collect and analyze telemetry data make it the go-to solution for network optimization tasks.

The ibnetdiscover utility is used to perform InfiniBand subnet discovery and outputs a human-readable topology file. GUIDs, node types, and port numbers are displayed, as well as port LIDs and node descriptions. All nodes and links are displayed, providing a full topology. This utility can also be used to list the current connected nodes. The output is printed to the standard output unless a topology file is specified.​

InfiniBand is a high-performance, low-latency interconnect technology used in AI and HPC data centers, particularly for east-west traffic between compute nodes in large-scale fabrics. Ensuring seamless data flow requires tools to troubleshoot and monitor the network, including the ability to trace and discover network paths between nodes. The question asks for the specific tool used to trace and discover paths in an InfiniBand fabric, which is a key task in InfiniBand troubleshooting.

According to NVIDIA’s official InfiniBand documentation, the ibnetdiscover tool is designed to discover and map the topology of an InfiniBand fabric, including the paths between nodes. It scans the fabric, queries the subnet manager, and generates a topology map that details the connections between switches, Host Channel Adapters (HCAs), and other devices. This tool is essential for verifying connectivity, identifying routing paths, and troubleshooting issues like misconfigured routes or link failures in large-scale InfiniBand fabrics.

Exact Extract from NVIDIA Documentation:

“The ibnetdiscover tool is used to discover the InfiniBand fabric topology and generate a map of the network. It queries the subnet manager to retrieve information about all nodes, switches, and links in the fabric, providing a detailed view of the paths between nodes. This tool is critical for troubleshooting connectivity issues and ensuring proper routing in InfiniBand networks.”

— NVIDIA InfiniBand Networking Guide

This extract confirms that ibnetdiscover is the correct tool for discovering network paths in an InfiniBand east-west fabric. It provides a comprehensive view of the fabric’s topology, enabling administrators to trace paths between compute nodes and ensure seamless data flow.

To check the status and link layer of InfiniBand interfaces, the ibstat command is used. For example:​

ibstat -d mlx5_0

This command provides detailed information about the InfiniBand device, including its state (e.g., Active), physical state (e.g., LinkUp), and link layer (e.g., InfiniBand).​

The correct utility is ibnetdiscover .

From the official NVIDIA InfiniBand Utilities Guide:

"ibnetdiscover scans the fabric and returns a topology of all switches and end nodes, including their GUIDs, LIDs, port numbers, and link speeds."

It generates a fabric map with node-to-port relationships and shows:

GUIDs

LIDs (Local IDs)

Link speeds and widths

Switch-to-host connections

This is essential for network topology validation and mapping physical port additions.

Incorrect Options:

ib_check_routes – for routing table diagnostics.

ibhosts – shows host information but not switch-level port mapping.

ibswitches – shows switch info, but lacks port-level LID/link speed mapping.

In InfiniBand networks, Partitioning using Partition Keys (PKeys) is the standard method for implementing multi-tenancy and traffic isolation. PKeys allow administrators to define logical partitions within the fabric, ensuring that traffic is confined to designated groups of nodes. This mechanism is essential for creating secure and isolated environments in multi-tenant architectures.​

The Unified Fabric Manager (UFM) leverages PKeys to manage these partitions effectively, enabling administrators to assign and control access rights across different tenants. This approach ensures that each tenant's traffic remains isolated, maintaining both security and performance integrity within the shared fabric.​

RoCE Adaptive Routing is a key feature of NVIDIA Spectrum-X that ensures high performance and resiliency in the network, even in the event of a link failure. This technology dynamically reroutes traffic to the least congested and operational paths, effectively mitigating the impact of link failures. By continuously evaluating the network's egress queue loads and receiving status notifications from neighboring switches, Spectrum-X can adaptively select optimal paths for data transmission. This ensures that the network maintains high throughput and low latency, crucial for AI workloads, even when certain links are down.​

Reference Extracts from NVIDIA Documentation:

"Spectrum-X employs global adaptive routing to quickly reroute traffic during link failures, minimizing disruptions and preserving optimal storage fabric utilization."​

"RoCE Adaptive Routing avoids congestion by dynamically routing large AI flows away from congestion points. This approach improves network resource utilization, leaf/spine efficiency, and performance."​

The NVIDIA UFM Cyber-AI Platform is specifically designed to enhance security and operational efficiency in InfiniBand data centers. It leverages AI-powered analytics to detect security threats, operational anomalies, and predict potential network failures. By analyzing real-time telemetry data, it identifies abnormal behaviors and performance degradation, enabling proactive maintenance and threat mitigation.​

This platform integrates with existing UFM Enterprise and Telemetry services to provide a comprehensive view of the network's health and security posture. It utilizes machine learning algorithms to establish baselines for normal operations and detect deviations that may indicate security breaches or hardware issues.​

Before upgrading Cumulus Linux, it's essential to back up configuration files to a different server. The /etc directory is the primary location for all configuration data in Cumulus Linux. Specifically, the following files and directories should be backed up:​

/etc/frr/ - Routing application (responsible for BGP and OSPF)

/etc/hostname - Configuration file for the hostname of the switch

/etc/network/ - Network configuration files, most notably /etc/network/interfaces and /etc/network/interfaces.d/

/etc/cumulus/acl - Access control list configurations​

Cumulus Linux is a network operating system used on NVIDIA Spectrum switches, including those in the Spectrum-X platform, to provide a Linux-based environment for Ethernet networking in AI and HPC data centers. When upgrading Cumulus Linux to a new version, it’s critical to migrate specific configuration files to preserve network settings and ensure continuity. The question asks for the two configuration file locations that should be migrated from the old installation during an upgrade.

According to NVIDIA’s official Cumulus Linux documentation, the key directories containing configuration files that should be migrated during an upgrade are /etc/cumulus/acl (for access control list configurations) and /etc/network (for network interface configurations). These directories store critical network settings that define the switch’s behavior, such as ACL rules and interface settings, which must be preserved to maintain network functionality after the upgrade.

Exact Extract from NVIDIA Documentation:

“When upgrading Cumulus Linux, you must back up and migrate specific configuration files to ensure continuity of network settings. The following directories should be included in the backup:

/etc/cumulus/acl: Contains access control list (ACL) configuration files that define packet filtering and security policies.

/etc/network: Contains network interface configuration files, such as interfaces and ifupdown2 settings, which define the network interfaces and their properties. Back up these directories before upgrading and restore them after the new version is installed to maintain consistent network behavior.” — NVIDIA Cumulus Linux Upgrade Guide

This extract confirms that options A and B are the correct answers, as /etc/cumulus/acl and /etc/network contain essential configuration files that must be migrated during a Cumulus Linux upgrade. These files ensure that ACL policies and network interface settings are preserved, which are critical for Spectrum-X configurations in AI networking environments.

In Cumulus Linux, microbursts are short-lived, high-volume traffic bursts that often go undetected by coarse-grained monitoring like SNMP.

The two tools specifically used for this purpose are:

What Just Happened (WJH)

"WJH provides real-time packet drop visibility and classifies drops by reason (e.g., congestion, ACLs, etc.), enabling microburst detection."

ASIC monitoring at millisecond granularity

"Deep telemetry is enabled via the switch ASIC, which provides sub-second counters that capture microburst patterns otherwise missed by SNMP."

Incorrect Options:

A and C provide low-frequency sampling, insufficient for microbursts which last milliseconds.

NVIDIA BlueField Data Processing Units (DPUs) are integral to securing AI infrastructures, especially in environments requiring stringent data privacy and security measures. BlueField DPUs offload and accelerate critical infrastructure tasks such as encryption, firewall enforcement, and intrusion detection, thereby isolating sensitive data paths from potential threats.​

In the context of AI workloads, BlueField DPUs enable secure and efficient data movement between GPUs and storage systems, ensuring that sensitive information, like financial data, is protected during both training and inference processes. Their integration into NVIDIA's reference architectures provides a hardware root of trust, essential for maintaining data integrity and compliance with security standards.​

In DPU Mode , the ARM cores on BlueField own the NIC data path , while still allowing the host system to access the DPU OS (via OOB or virtio) .

From NVIDIA BlueField Documentation:

"In DPU Mode, the data path is offloaded to the BlueField Arm cores, enabling advanced security and networking functions, while still allowing host access to the BlueField OS."

This is different from:

NIC Mode : Data path controlled by host, ARM cores inactive.

Separated Host Mode : Complete isolation; host cannot access DPU OS.

Restricted Mode : Limited host access to DPU OS, but without full offload capabilities.