G1OldCSetRegionThresholdPercent under ExperimentalFlag

Thomas Schatzl thomas.schatzl at oracle.com
Thu Jun 22 10:18:19 UTC 2017


On Tue, 2017-06-20 at 23:45 -0700, Sundara Mohan M wrote:
> Hi,
>   Can someone shed more light on why G1OldCSetRegionThresholdPercent
> flag is under experimental (Need to add  -
> XX:+UnlockExperimentalVMOptions to modify it.)

  in my view -XX:+UnlockExperimentalVMOptions mostly serves mostly as a
"I really want to do that and I know what I am doing" confirmation from
the user that he is aware that using this (in this case) option to
influence the set of regions taken in during mixed gc you might get
surprising behavior.

Also, I think there has been no official documentation for it - also
because it should be very rarely needed.
In particular, I am curious about the case when it would be useful to
change it. Could you give some log files showing that there is an issue
with the upper bound for the number of old gen regions to take during
GC? (i.e. the amount of old gen regions taken is too small and there is
ample pause time left and it matters to clean up more regions in a
single mixed gc?)

Sometimes there are problems with the lower bound that is controlled by
the -XX:G1MixedGCCountTarget (product level) option.


More information about the hotspot-gc-dev mailing list