RFR: JDK-8262068: Improve G1 Full GC by skipping compaction for regions with high survival ratio [v8]

[Thomas Schatzl]

Hi,

On 12.03.21 06:13, Hamlin Li wrote:
> On Thu, 11 Mar 2021 15:19:36 GMT, Thomas Schatzl <tschatzl at openjdk.org> wrote:
> 
>>> Hamlin Li has updated the pull request incrementally with one additional commit since the last revision:
>>>
>>>    fix bot crash.
>>
>> src/hotspot/share/gc/g1/g1FullGCMarker.hpp line 104:
>>
>>> 102:
>>> 103:   void flush_mark_region_cache() {
>>> 104:     if (MarkSweepDeadRatio > 0) {
>>
>> Drop this check to make the code more straightforward. Other code like in PR #2579 might find this information useful too. Maybe this could even be factored out in a separate CR.
> 
> Hi Thomas,
> Do you mean to collect liveness info in mark phase when G1 full gc, even if MarkSweepDeadRatio is == 0 (which means do not skip any regions when compaction)? If this is the request, I can file a new bug for this liveness collection in G1 full gc.

Yes, please split out the liveness info gathering in the mark phase into 
a separate CR. This is what PR #2579 needs.
StefanJ and me were discussing adding this already when implementing the 
G1 parallel full gc, but due to lack of users we refrained from that..

> For the jfr PR #2579 , I think it's can be addressed in a separate bug which depends on the one for liveness collection.
> I will file/work these 2 new bugs (liveness collection when g1 full gc, jfr liveness event after full gc) if this is what you suggested.

PR #2579 is about the JFR liveness event. Jbachorik (the author of that 
PR) could just reuse the information gathered in that new CR then. 
Please coordinate with him if needed, but I do not think you 
specifically needs to worry about the JFR event, I assume that jbachorik 
will likely be happy to be able to add the plumbing for the JFR event then.

I cc'ed him (using some email address from the jfr-dev mailing list, I 
hope I got it right).

Thanks,
   Thomas