THE TIME CONSUMPTION FROM THE TSEC TO THE CORE THROUGH DPAA

[fangxiao]

Is there any reference about the delay from TSEC to the core? I know that the accurate time consumption varies from the number of the existing work queues and the tasks which DPAA must do. However, it is useful to know the approximate level of delay when a single package (there is no other tasks in DPAA) is transmitted from the input of the TSEC to the e500 core, because in our communication system this delay is a sensitive parameter. It decides that whether a package can be delivered to the e500 core from the ethernet before the deadline time.

[lunminliang]

See these data from helpdesk database:

All the latency number provided here are measured by Spirent traffic generator. Software optimization and the newer release of SDK will improve the result.

1.With the ball -> Processor (linux swap the src and dest IP addr) -> ball, Spirent reports min latency for ~2.65us

2.For simple IP forwarding, Spirent measures round trip min latency for ~6us

3. ball -> Processor latency is in the range of half of answer 1, minus the software overhead.