Reproducing network performance benchmark

I am testing the network performance on the Ten64 and I would like to reproduce the results of your previous tests in Network Performance - Traverse Ten64 Documentation so I could both verify that my test environment looks reasonable and so I can experiment with various alternatives, such as DPDK, to see how that improves (or maybe doesn’t improve) network performance.

The performance document mentions using ls-addni to configure the 10G ports. So I can make sure I am comparing my numbers to your numbers fairly, do you still have the ls-addni commands you used? And, is there any other configuration I should adhere to?

The requirement to start with the “single port” configuration and add the required ports manually should not be needed anymore, it was due to a difficulty getting the DPAA2 to balance flows across all the CPU cores with all 10 ports enabled.

That said, you might find you can still tune the configuration to your use case. e.g if you know a port is being used as a L3 WAN only you could provide less MAC filter slots vs one being used for LAN.

Have a look at this thread, it provides details on how to “start from scratch” and later details of the fix:

Also, kernel 5.16 onwards added IRQ coalescing, which should improve performance a bit. We haven’t done comprehensive benchmarks on this yet.
https://patchwork.kernel.org/project/netdevbpf/cover/20211015090127.241910-1-ioana.ciornei@nxp.com/