It would be nice to see how much time a packet is spending during DataPlane pipeline processing, which could be useful for analyzing performance issues and/or optimizations.
Since the analysis should not result in performance degradation, I would suggest the following approach:
Create a shared memory segment
Allocate a dedicated memory block for each worker to avoid interference between them.
Map the entire block to a sequence of counters initialized to zero.
Get the TSC counter between each pipeline step we want to analyze.
Increasing the corresponding counter for each employee by the difference between the TCS counters immediately after and before the analysis stages.
After this we could create a reader application that does:
Attach to shared memory segment
Read counters for workers in a loop
Show the difference between the next and previous values
It would be nice to see how much time a packet is spending during
DataPlanepipeline processing, which could be useful for analyzing performance issues and/or optimizations.Since the analysis should not result in performance degradation, I would suggest the following approach:
TSCcounter between each pipeline step we want to analyze.TCScounters immediately after and before the analysis stages.After this we could create a reader application that does: