RFR: 8319048: Monitor deflation unlink phase prolongs time to safepoint

[Aleksey Shipilev]

On Mon, 6 Nov 2023 05:50:54 GMT, David Holmes <dholmes at openjdk.org> wrote:

> FWIW I see for baseline versus patched:
> 
> ```
> 77.73± 1.65 | 122.95± 44.39  p = 0.001
> ```

The error margin on the "patched" looks way too high. Even naively, 122.95 - 44.39 = 78.56, and 77.73 + 1.65 = 79.38, so the confidence intervals intersect. I cannot see right away what could explain more variability in patched version, as the test does _not_ involve monitors at all, AFAICS.

> but the results you pasted, if I am reading them right, indicate the baseline is ~2x better than patched: 2ns/op versus 4ns/op. ???

That's the error bars. The actual diff is (68.8 +- 2.2) vs (69.3 +- 4.1) ns/op.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/16412#issuecomment-1794502039