The author mistakes increasing concurrent request as being some sort of proxy for concurrent users. Once you have maximized rps or maximized cpu, increasing con

The author mistakes increasing concurrent request as being some sort of proxy for concurrent users. Once you have maximized rps or maximized cpu, increasing concurrent just artificially hurts your average latency. Max RPS equals number of users you can serve per second. Concurrency should just be tweaked until you eliminate the network latency of your test framework.