On Thu, 10 Aug 2023 11:40:52 GMT, Pavel Rappo <pra...@openjdk.org> wrote:
>> My bad. Forget my note > >> Some have a preference for providing a seed for `Random` instances in >> micros. Either hard-coded or through a `@Param` (I find this a bit >> excessive). Doing so might reduce run-to-run noise. > > I hear you, but unless your opinion is strong, I'll leave it as is. Hard-coding a seed would remove the run-to-run randomness, too, so would remove the need the `randomness` tag if this utility were to be used from a functional test. But yes, the reason to not do true randomness when setting up data for micros is to ensure we don't get multimodal results from arbitrary differences in the data setup. There are other ways around that, e.g. randomize data continually between iterations, run more forks to get more samples about flicker etc. A lot of work with an end result which is unlikely to add little more quality than a "random" set of data that we keep static from run-to-run. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/14630#discussion_r1289995785