JEP 248: Make G1 the Default Garbage Collector
Yu Zhang
yu.zhang at oracle.com
Tue Jun 2 18:33:07 UTC 2015
sorry if this is a duplicate message. I had some problems sending it
earlier.
The correlation between young gc pause and occupancy in tenure, might be
related to Remember Set operations. Do you see ScanRS/UpdateRS goes up
with the tenure occupancy is high?
If the occupancy of a region in old gen is higher than
G1MixedLiveThresholdPercent(default 65 before 8u40, 85 8u40 and later),
then it is not considered as a candidate for cset.
Thanks,
Jenny
On 6/2/2015 5:19 AM, Kirk Pepperdine wrote:
> Hi,
>
> I can show you a direct correlation between young gen pauses and
> occupancy of Tenured. In fact please find attached graphs from work I
> did on Monday. What I do know is that under normal operations there
> are regions that I wished could be collected that can’t be collected.
> I say this because they don’t meet the thresholds to be included in
> the CSet. That said, the amount of dead in each of the tenured regions
> is enough that overall there is a visible negative effect on GC pause
> times. So, I think I'd like a continuous concurrent cleaning of
> tenured regions or something else that I’ve not quite thought through
> just yet. What I do know is that if I run a full gc, the pause time
> picture clears up because the regions that wouldn’t get collected do
> and that reduces the “live set”. Some where tucked away I’ve got
> others logs that demonstrate this effect. I’ll see if I can dig it up.
>
> Regards,
> Kirk
> PS, once we’re finished with this conversation I’d like to tackle
> safe-pointing behavior. I think Charlie knows what I’m talking about.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-gc-dev/attachments/20150602/ee3a21d8/attachment.htm>
More information about the hotspot-gc-dev
mailing list