Withdrawn: 7903368: JMH: GC profiler misreports allocation and churn rates
Aleksey Shipilev
shade at openjdk.org
Wed Oct 19 17:51:23 UTC 2022
On Wed, 19 Oct 2022 15:33:40 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:
> Look at the new test. If you multiply the allocation rate from `gc.alloc.rate.norm` by the benchmark throughput, it would disagree with `gc.alloc.rate`. This is because `gc.alloc.rate` is computed by GC profiler using its own timer, which accidentally includes the artificial delay before deregistering the GC notifications.
>
> The current delay is 500ms. With 1000ms iterations, the disagreement would be about (1000+500)/1000 = 1.5x!
This pull request has been closed without being integrated.
-------------
PR: https://git.openjdk.org/jmh/pull/87
More information about the jmh-dev
mailing list