In an NVIDIA InfiniBand fabric managed by the Unified Fabric Manager (UFM), you need to collect local telemetry data, monitor performance statistics directly from the compute nodes, and orchestrate remote firmware upgrades. Which component must be running on the compute nodes to perform these actions?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal with managing high-speed InfiniBand fabrics: you need eyes and ears on the ground. In a UFM setup, the UFM Host Agent is the software component that actually runs directly on your compute nodes. It gathers local node health, checks performance stats, and handles tasks like updating OFED drivers and firmware remotely. The UFM Server is the brain of the operation, coordinating the whole fabric. The UFM Switch Agent sits on the physical switches to manage them, and the UFM-SDN Appliance is the dedicated hardware box running the UFM platform itself. So, if it's node-level monitoring and upgrades you need, the Host Agent is your answer.
Full explanation below image
Full Explanation
The NVIDIA Unified Fabric Manager (UFM) platform is used to manage, monitor, and optimize InfiniBand networking fabrics within scale-out AI and HPC clusters. The architecture of UFM relies on distinct software components to interact with different layers of the infrastructure. The UFM Host Agent is a lightweight daemon installed directly on the compute hosts (servers) in the fabric. It is responsible for gathering node-specific telemetry, monitoring port status and local host performance metrics, and executing administrative commands sent from the central manager, such as updating Host Channel Adapter (HCA) firmware or Mellanox OpenFabrics Enterprise Distribution (OFED) drivers. The other components serve different functions within the management framework. The UFM Server is the central management engine that runs on a dedicated controller or server, providing network topology discovery, path calculation, routing algorithms, and overall fabric orchestration. The UFM Switch Agent runs locally on managed InfiniBand switches to assist with switch discovery, configuration, and monitoring. The UFM-SDN Appliance is a pre-configured physical server appliance designed to run the central UFM software suite, providing a turn-key solution for network administrators. Only the UFM Host Agent operates on the compute nodes to gather local telemetry and execute node-level maintenance tasks, making it the correct choice.