On Thu, 18 Nov 2021 11:38:08 GMT, Andrew Leonard <aleon...@openjdk.org> wrote:
>> Just to verify, this test jar had 16000 files in a single directory? Since >> having 100 files in 160 directories would not have the same impact, right? > >> Just to verify, this test jar had 16000 files in a single directory? Since >> having 100 files in 160 directories would not have the same impact, right? > > @magicus so the first test I did was 16,000 files in multiple directories, as > I thought that was more representative, and would involve multiple small > sorts vs one large sort. > I have just re-run the tests again, for both 16000 multi-dir and 16000 single > dir, and the results are similar: > 16000 multi-dir x50: > Existing code: 291 seconds (variance: +/- 3 seconds) > New patch: 293 seconds (variance: +/- 3 seconds) > 16000 single-dir x50: > Existing code: 235 seconds (variance: +/- 3 seconds) > New patch: 237 seconds (variance: +/- 3 seconds) @andrew-m-leonard Thanks. I think that alleviates any fear of performance regression. Sounds like you are well within the span of normal variation. ------------- PR: https://git.openjdk.java.net/jdk/pull/6395
