RFR: 8322537: Parallel: Remove experimental adjustment in PSAdaptiveSizePolicy

[Thomas Schatzl]

On Tue, 2 Jan 2024 09:32:21 GMT, Albert Mingkun Yang <ayang at openjdk.org> wrote:

>> src/hotspot/share/gc/parallel/psAdaptiveSizePolicy.cpp line 591:
>> 
>>> 589:     // *desired_eden_size_ptr = *desired_eden_size_ptr + eden_heap_delta;
>>> 590:     set_change_young_gen_for_min_pauses(
>>> 591:       increase_young_gen_for_min_pauses_true);
>> 
>> Removing this setter (and other similar ones) changes the values of the externally observable performance counters. Not sure if this has been intended. I suggest keeping removng this "EXPERIMENTAL ADJUST" comment separate from externally visible changes.
>> 
>> The "    // Only record that the estimator indicated such an action." part of the comment also seems not to be part of the "experimental adjustment" comment.
>
>> Removing this setter (and other similar ones) changes the values of the externally observable performance counters. Not sure if this has been intended.
> 
> Yes, that is intended. That perf-counter lives in the "unstable and unsupported name space" (https://github.com/openjdk/jdk/blob/master/src/hotspot/share/runtime/perfData.cpp#L56).
> 
> Having the indicated action recorded in the perf-counter was probably used to aid the development of AdaptiveSizePolicy. However, its existence solely complicates the flow nowadays. Therefore, I propose removing it completely. (Ofc, one can always perform another round of experiment to improve AdaptiveSizePolicy afterwards.)

I am okay with removing the counter after a quick search if it seems to be used, but if so I would prefer to remove it completely not only the counter updates.
That at least sends a clear signal that the counter is not available any more to anyone actually using it; silently returning garbage seems worse service to users.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/17166#discussion_r1448514294