RFR: JDK-8151674: STW phases at Concurrent GC should count in PerfCounter

Yasumasa Suenaga yasuenag at gmail.com
Fri Mar 18 14:14:44 UTC 2016


Hi,

> I added a comment to JBS about G1 initial mark.
> I will upload new webrev after discussing for it.

We discussed about it on JBS, and we decided not to count initial mark on G1
to CGC counter. [1]

I uploaded new webrev. Could you review it?

   hotspot: http://cr.openjdk.java.net/~ysuenaga/JDK-8151674/webrev.01/hotspot/
       jdk: http://cr.openjdk.java.net/~ysuenaga/JDK-8151674/webrev.01/jdk/

jdk part is the same with the patch which is reviewed by Staffan Larsen. [2]


Thanks,

Yasumasa


[1] https://bugs.openjdk.java.net/browse/JDK-8151674?focusedCommentId=13914167&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13914167
[2] http://mail.openjdk.java.net/pipermail/hotspot-gc-dev/2016-March/016911.html


On 2016/03/17 8:23, Yasumasa Suenaga wrote:
> Hi Jon,
>
> 2016/03/17 3:25 "Jon Masamitsu" <jon.masamitsu at oracle.com <mailto:jon.masamitsu at oracle.com>>:
>  >
>  >
>  >
>  > On 03/10/2016 07:59 PM, Yasumasa Suenaga wrote:
>  >>
>  >> Hi all,
>  >>
>  >> This review request continues from:
>  >> http://mail.openjdk.java.net/pipermail/hotspot-gc-dev/2016-March/016774.html
>  >> http://mail.openjdk.java.net/pipermail/hotspot-gc-dev/2016-March/016896.html
>  >>
>  >> I wonder that STW phases (Remark and Cleanup) at G1 are not counted in jstat FGC column.
>  >> For example, Initial Mark and Remark at CMS are counted as FGC.
>  >> We discussed on hotspot-gc-dev for this issue, and I proposed to add new PerfCounter
>  >> for CGC STW phases.
>  >>
>  >> I uploaded webrev. Could you review it?
>  >>      hotspot: http://cr.openjdk.java.net/~ysuenaga/JDK-8151674/webrev.00/hotspot/
>  >>          jdk: http://cr.openjdk.java.net/~ysuenaga/JDK-8151674/webrev.00/jdk/
>  >>
>  >> For compatibility, this patch works the same as the current by default.
>  >> If you set -XX:+EnableConcGCPerfCounter, CGC counter will work fine.
>  >> (I want to set +EnableConcGCPerfCounter by default)
>  >
>  >
>  > I would prefer not to have the flag (EnableConcGCPerfCounte)
>  > and just make the change for CMS.  This is a change for a major
>  > release and I think we can accept the incompatibility.
>
> Okay, I will remove this flag.
>
> I added a comment to JBS about G1 initial mark.
> I will upload new webrev after discussing for it.
>
>  > http://cr.openjdk.java.net/~ysuenaga/JDK-8151674/webrev.00/hotspot/src/share/vm/gc/g1/g1MonitoringSupport.cpp.udiff.html
>  >
>  > + // name "collector.2". In a generational collector this would be the
>  > + // STW phases in concurrent collection.
>  > + _conc_collection_counters =
>  > + new CollectorCounters("G1 stop-the-world phases", 2);
>  >
>  >
>  > Is there a hard connection between these phases and a generational
>  > collector (i.e., should the "In a generational collector this would be the"
>  > be dropped in the comment)?
>  >
>  > A comment such as
>  >
>  > // name "collector.2". STW phases as part of a concurrent collection.
>
> Thanks, I will fix.
>
>  > A set of JDK changes had been previously been reviewed (not by me). Are these
>  > the same?
>  >
>  > http://cr.openjdk.java.net/~ysuenaga/JDK-8151674/webrev.00/jdk/
>
> Yes, same patch.
>
> Yasumasa
>
>  > Jon
>  >
>  >
>  >
>  >>
>  >>
>  >> I cannot access JPRT.
>  >> So I need a sponsor.
>  >>
>  >>
>  >> Thanks,
>  >>
>  >> Yasumasa
>  >
>  >
>


More information about the hotspot-gc-dev mailing list