We have an application that has a requirement for our users to throughput x no. of transactions in a given period. Taking into account expected server time the user time left to perform the transaction is sub 1 second. Also in another area we want to look at how the server perfoms if the users increase their productivity by say 25% which gets us into fractional seconds.
QALoad will measure the fractional seconds when you are running the tests. So you know how long each transaction is taking down to the millisecond.
I'm not clear where fractional seconds of sleep comes in when you want to measure fractional seconds on performance. I don't see the two as related. Sleep is not peformance.
If you want to increase "user productivity" by 25%, instead of dialing down sleep, why not just add 25% more VUs without changing anything else? Won't it have the same effect? I'm not sure that you will see any difference by changing fractional seconds of sleep. In my experience, you don't see much difference unless you go from no sleep to sleep, or change sleep by large amounts, say from really pounding the server to sending 1 transaction per 10 seconds.
If you are really interested in seeing how the server performs, turn sleep off. Then you will be pounding the server with the maximum rate of traffic.
Thanks for the advice - I'd thought about altering the sleep factor % as a way around this - however the DO_MSLEEP (millisecond sleep)looks a far more refined approach and should allow me to get the fractional second sleep I was after.