-XX:+ParallelRefProcEnabled may make it worse

[Nat]

Recently, I found a situation where -XX:+ParallelRefProcEnabled makes the GC time to be significantly worse, especially the tail end. It seems to be much worse with multiple core system. I believe it is a bug that might not be addressed yet. The key problem seems to be around the task queue implementation rather than the actual work itself. So instead of just enabling it blindly, make sure you cut and measure. 

I will update if I find more information.

[Kodewerk]

Hi,

You can log reference processing counts and times… If you’re at about 50K references/second then using parallel reference processing starts to make sense… prior, single threaded is better.

Kind regards,

Kirk

On Jan 29, 2019, at 8:56 AM, Nat <nat....@gmail.com> wrote:

Recently, I found a situation where -XX:+ParallelRefProcEnabled makes the GC time to be significantly worse, especially the tail end. It seems to be much worse with multiple core system. I believe it is a bug that might not be addressed yet. The key problem seems to be around the task queue implementation rather than the actual work itself. So instead of just enabling it blindly, make sure you cut and measure. 

I will update if I find more information.

--
You received this message because you are subscribed to the Google Groups "mechanical-sympathy" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mechanical-symp...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.