Generate Load by adding reasonable think time

I used to generate load by launching small number of VUs without any think time during performance testing. The pros is You can trigger a very high load/Throughput even you are not creating many concurrent threads, but the cons is you can not generate consistent/stable load to make comparison once you make some tuning or changes. Recently, I am discussing with my teammate ZhouZhou, and figure out how we generate a consistent load by adding reasonable think time(Little Law helps the calculation here) between each request, here is the diagram we made to show how much load/Throughput we can generate/get: