You are administering an InfiniBand network fabric managed by the NVIDIA Unified Fabric Manager (UFM). To monitor local node metrics, collect system performance statistics, and push remote firmware updates directly to the Host Channel Adapters (HCAs) on individual servers, which component must be running on the compute nodes?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Okay, let's dive into UFM. Imagine your boss walks in and asks you to upgrade the HCA firmware across fifty compute nodes by the end of the day. You don't want to log into each box manually, right? That's where the UFM Host Agent comes to the rescue. It's a lightweight daemon you run directly on the host OS of your servers. It talks back to the central UFM server, sharing CPU, memory, and HCA stats, and allows you to push those firmware updates remotely. UFM Switch Agent is for managed switches, and OpenSM is the subnet manager that builds the routing tables. Got it? The Host Agent is your agent on the server side.
Full explanation below image
Full Explanation
The NVIDIA Unified Fabric Manager (UFM) platform is an end-to-end fabric management solution for InfiniBand networks. It consists of multiple components distributed across the fabric to enable monitoring, routing optimization, and device management.
The UFM Host Agent is an optional software component installed directly on the operating system of the compute nodes (servers) connected to the fabric. It acts as a local proxy that performs the following key functions: 1. Discovery: It gathers host-specific parameters, such as host name, OS version, CPU details, and PCI slot mappings, which cannot be discovered solely via in-band network sweeps. 2. Performance Monitoring: It collects local telemetry data and hardware performance statistics, sending them to the central UFM system. 3. Device Management: It facilitates remote maintenance tasks, including upgrading Host Channel Adapter (HCA) firmware without requiring manual SSH access to each node.
Let's look at why the other options are incorrect: - UFM Switch Agent (Option B) runs locally on managed InfiniBand switches to assist UFM in managing switch hardware parameters and status, not host nodes. - UFM Telemetry Platform (Option C) is the centralized management server that aggregates data and handles routing logic but doesn't run on the compute nodes directly to perform local host tasks. - Open Subnet Manager (OpenSM) (Option D) is the subnet manager responsible for discovering the network topology, assigning Local Identifiers (LIDs), and configuring routing tables. It does not manage host-level system parameters or upgrade HCA firmware.