I believe my original solution was something closer to what you are suggesting for backward compatibility. I originally had a flag that enabled changing source port instead of source IP addresses, but I received feedback that adding an extra flag was complicating things too much from Stephen.

On a VM, the purpose of using multi-flow is similar to that of bare metal: to test RSS in the RX side. However, generating traffic by changing source IP address can cause inconsistencies in performance due to protections in cloud infrastructure from sending packets from a different source IP address than is provisioned for the VM. Changing source UDP port to test RSS should be functionally equivalent while allowing VMs to send traffic from a single source IP address.

If everyone agrees that adding --txonly-multi-flow as an option as well as keeping the flag is an acceptable way of moving forward, I can do that.

Thanks,

Josh