Telemetry is the process of collecting, transmitting, and analyzing data from remote devices or systems. Telemetry can provide valuable insights into the performance, health, and behavior of the devices or systems, as well as the environment and context in which they operate.
Switches can generate and consume a large amount of data, such as traffic statistics, port status, device configuration, device health, and events. This data can be used for telemetry purposes, such as monitoring the switch performance, health, and behavior, and the network topology, traffic patterns, and quality of service.
It is used to monitor InfiniBand fabric statistics such as bandwidth, congestion, errors, and latency for real-time monitoring and historical analysis.
A user can choose whether to fetch it from each InfiniBand switch which runs gNMI (gRPC Network Management Interface) server as part of NVOS or while using the central UFM appliance.
NVOS gNMI is better for high-scale, low-latency streaming telemetry, while UFM is better for fabric management and minutes-based monitoring.
A practical rule of thumb from the current XDR doc set is: if there is a data lake / continuous telemetry pipeline, use gNMI; otherwise use UFM.
How to configure gNMI agent can be found in each NVOS User Manual, starting from NVOS Version 25.03.1010 under “Telemetry Streaming”.
Last updated: