Layer one: devices and links
Check model identity, firmware, drivers, port rate, and management reachability. A link indicator confirms only part of the required behavior. Keep a port-to-cable map so faults can be traced to a specific path.
Layer two: sustained operation
Observe errors, retransmissions, loss, congestion, and temperature through continuous monitoring. Use validation methods appropriate to RoCE or InfiniBand and the deployed equipment. Record versions and conditions for each baseline.
Layer three: workloads and recovery
Run representative communication and application traffic, then examine scaling and recovery behavior. Distinguish connectivity checks, laboratory tests, and business workload results. Publish comparisons only when measurements use comparable conditions.
Related products
These products are topic references, not a certified configuration. Verify exact part numbers and deployed software.
- NVIDIA MQM9700-NS2F 400G per NDR port InfiniBand Switch
- NVIDIA SN5400, Power-to-Connector airflow 920-9N42C-00FB-7C0 400GbE 2U Ethernet Switch
- NVIDIA MCX75343AMS-NEAC 400GbE OCP3.0 Network Adapter
- NVIDIA MUA9702H-2SFS 1U Network Appliance
