RFR: 8318190: GenShen: Be less aggressive with triggers for old-gen has expanded [v3]

[Kelvin Nilsen]

On Tue, 17 Oct 2023 23:54:27 GMT, William Kemper <wkemper at openjdk.org> wrote:

>> Just to be sure: are you suggesting that before we raise the _growth_trigger, we should confirm that we are not doing mixed evacuations?  And then we don't have to recheck whether the trigger condition is still valid when we get to here?
>
> I was suggesting we move [the logic that raises these triggers](https://github.com/openjdk/shenandoah/blob/master/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp#L3100-L3123) into the heuristic itself. This would reduce changes in shenandoahHeap.cpp and better encapsulate the logic for the heuristic.
> 
> If we want to align evaluation of these triggers with rebuilding the freeset (so that they aren't being evaluated at the same frequency as the young generation), we could add a `heap_has_changed` flag (or something). I'm also working on a change to only evaluate old generation triggers when the young generation trigger has fired.

Thanks for clarifying.  I "guess" the main tradeoff is how often do we do the work of evaluating the trigger condition.  As currently implemented, we evaluate this condition at the end of each GC cycle, which is relatively rare.  If we move it to should_start_gc(), it could be revaluated every ms or so.  On the other hand, after we integrate your change, the question of should_start_gc(OLD) will only be asked at the start of a new young gc, which is also relatively rare.

I'll move it now, in anticipation of your improvement, if you concur.

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

PR Review Comment: https://git.openjdk.org/shenandoah/pull/341#discussion_r1364515072