Hi @cyberpank
Have you tried profiling with different memory and CPU configurations for the brokers?
You are measuring the end-to-end processing time of each process, I take it.
I would measure the theoretical minimum E2E of the process by running one worker for each task, and starting one process and measuring the E2E time. It is never going to be faster than this.
Then try ramping up concurrent requests, and see how the time changes as the numbers ramp up.
Then see what effect CPU and memory resources have.
See also these posts:
Josh