RFR: 8030815: Code roots are not accounted for in region prediction [v2]

[Thomas Schatzl]

On Thu, 14 Sep 2023 13:32:23 GMT, Albert Mingkun Yang <ayang at openjdk.org> wrote:

> > However some recent investigations showed that it can be of significant importance.
> 
> What/how much is the effect of this better prediction when the code-root time is significant?

With  the current code root set implementation it is not measurable - it's too erratic due to imbalances in code root scan and general locking all around to be able to usefully measure.
>From what I saw with [JDK-8315503](https://bugs.openjdk.org/browse/JDK-8315503), some pause time peaks with some stress test were removed (even after parallelization/better balancing you get some significant code root scan times up to 40ms on average).

The reason why this change comes first (and not e.g. the mentioned JDK-8315503) is that this change is independent of other changes related to class unloading that JDK-8315503 requires. I needed to start somewhere with changes.

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

PR Comment: https://git.openjdk.org/jdk/pull/15632#issuecomment-1723141655