JMH branch testing

[Bernd Eckenfels]

Hello,

looking for some ideas. I want to benchmark different implementations
for a "remeber maximum value" statistics function (which is used multi
threaded). One behaviour is of course, that the performance might
differ for numbers which are and which are not higher than the current
maximum. And then also the impact of varrying branches.

What I currently have is not very nice, as the distribution of the
random numbers (and the creation of them) does not execise real world
cases. Before I spent some time having a better distribution I wonder
if using a deterministic "measure <currentmax and >currentmax cases"
solution would be better or worse?

https://gist.github.com/ecki/8350a276caa444a62dce

Gruss
Bernd