nambar wrote:But if performance is critical, then you need some kind of performance tests anway. At least to find performance regressions introduced by new code.
Simple testing doesn't show real impact and we can't get to real life load in our testing.
We'd like to know if the JVM is actually doing anything different (after initialization) when running with these flagsI don't know.
nambar wrote:That's a big difference, but a bad test. You should really be running your production code in a performance test.
A simple test (for loop, after JVM was initialized) does show ~40% hit in run time.