Open baiwfg2 opened 4 years ago
Check the following two images. Why does the latency go up heavily when I increase rate ?
When I want rate to be fixed, but I don't know the lower bound numbers of connections can reach that fixed rate. After all I'd rather choose that number, because when num of connections go up, the latency go up too.
Just some thoughts, this doesn't answer your question.
I think you bring up a great question. Some general guidance on how to set these parameters would be great. That said, I think the answer to your question of "what if I want an unlimited rate" is to simply set rate some ungodly large number like 10 billion requests per second. Of course, what this means is that whatever latency is reported is not useful.
As for connection count. I think the appropriate value here might be more related to the server you are stress/load testing. It's likely dependent on how many connections the server you are trying to test can handle. Also, don't set it so large that you see any sort of connection/socket errors.
As for threads, I always just match it up with the number of threads/cores on the machine running wrk2.
Greate qustion.
Check the following two images. The result is very confusing!
These three factors are tightly related. It's very easy for users to give a wrong rate which doesn't match with the given threads or connections. How should we avoid this ?
Furthermore I don't understand why rate parameter is required. What if I just want to stress the server with unlimited rate ?