Repeatability of performance results
I have a problem with repeatability of performance test results.
I'm using Japex to test performance of Calendar.getInstance() method.
run time = 10 seconds,
number of threads = 10.
The results are:
first launch = 178643tps
second launch = 188834tps
third launch = 189533tps.
So optimum evasion is 2%.
Is it possible to get more repeatable results?
Thanks in advance.
Re: Repeatability of performance results
The accuracy of the PC clock is limited, plus background processes can influence the performance results, too.
I would run a longer test (say 60-300 seconds) to reduce the variation.