What Is Port-to-Port Latency?_
The value aggregates serialization, forwarding lookup, and any queuing delay inside the device. In cut-through mode the switch begins transmission before the full frame arrives, lowering the contribution of serialization. In store-and-forward mode the entire frame must be buffered before egress begins.
Technical Details
Reported figures are normally the minimum latency observed under zero load. Actual latency rises with queue depth when the egress port is contended. Measurement is performed with hardware timestamping at both ports or with a dedicated latency tester; software-based pings are unsuitable for fabric characterization.
How Leviathan Systems Works with Port-to-Port Latency
During commissioning the field crew records port-to-port latency on the InfiniBand or Ethernet leaf and spine switches that form the GPU scale-out fabric. Results are compared with the OEM specification before the cluster is released for workload testing.
Related service
Network Testing & Commissioning →