RFR: 8294850: Make rs length/pending card predictors dependent on gc phase

Kim Barrett kbarrett at openjdk.org
Tue Oct 11 23:21:10 UTC 2022


On Tue, 11 Oct 2022 12:53:29 GMT, Thomas Schatzl <tschatzl at openjdk.org> wrote:

> Hi all,
> 
>   can I have reviews for this change that separates `rs_length` and `pending_cards` predictors by young phase?
> 
> Some notes:
> * this does not intend to improve the predictors themselves - i.e. the use of `xxx_diff` and `xxx` predictors and adding them together is kind of weird, imho trying to add a kind of prediction compensation/improvement to act on differences like a (PID-)controller. I also strongly suspect that the `xxx_diff` predictors ought to be allowed to report negative values. This is no attempt to fix this
> * the parameter `for_young_gc` in the analytics API is misnamed. It's not about young gcs, but young_only_phase - I will fix this in a follow-up
> 
> Depends on PR#10649, please review that one first.
> 
> Thanks, 
>   Thomas
> 
> Testing: gha, local testing, tier1-5 with other changes in this patch series, perf testing with other changes in this patch series

Looks good.

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

Marked as reviewed by kbarrett (Reviewer).

PR: https://git.openjdk.org/jdk/pull/10650



More information about the hotspot-gc-dev mailing list