RFR: 8256265 G1: Improve parallelism in regions that failed evacuation [v2]

[Hamlin Li]

On Fri, 3 Dec 2021 12:25:40 GMT, Thomas Schatzl <tschatzl at openjdk.org> wrote:

> It would also be very interesting to get information to differentiate how long the sorting and how long the actual self-forward removal of that phase took. Additional interesting information is memory consumption of the segment allocators/sorting. If significant compared to other allocation, maybe we could even separate them out in a separate category (not sure, I do not really think so, just thinking out loud).

Agree, these information will be very helpful to the subsequent optimizations we will soon work on.

> 
> If you need help with adding gc pause log statistics, ping me; 

Thanks a lot for your help all the time. :) I will ask for your help.

> for timing information it should be sufficient to add enum values to `G1GCPhaseTimes::GCParPhases`, and initialize them like the others at the appropriate places; counts can be added as "thread work item" like for the `ScanHR` enum value.

Thanks for the information. :)

> 
> The current evacuation failure handling implementation is already really good compared to 17 due to your hard work, now I guess it starts to become necessary knowing more details for further investigation and in general, for analysis and troubleshooting by users now and later :-).

Agree, we need to supply more detailed information to users.

> 
> Thanks a lot for your effort, Thomas

Thanks a lot, my pleasure :)

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

PR: https://git.openjdk.java.net/jdk/pull/6627