You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@harmony.apache.org by Xiao-Feng Li <xi...@gmail.com> on 2008/03/20 15:23:42 UTC
[drlvm][gc] switch the default GC minor collection to semi-space algorithm
Hello, since the semi-space algortithm recently developed in Harmony
GC has obviously better performance than the current partial-forward
algorithm when they are used for minor collection, I plan to switch
the default minor collection to semi-space algorithm. The results in
my runs have no new failures with classlib unit tests, drlvm tests and
reg tests, compared to current default GC on Linux/Win 32/64-bit
platforms. Although I have strong confidence that the failure cases
should be minimum, I still expect there be some regressions with other
untested test cases. I hope the semispace GC to be the default in next
milestone build. Please forgive me if the switch causes some
unexpected failures, and GC people will try best to fix the failures
ASAP.
Please let me know if anybody has objections. If no objections, I will
make the switch in 12 hours. Thanks for your supports.
Thanks,
xiaofeng
--
http://xiao-feng.blogspot.com
Re: [drlvm][gc] switch the default GC minor collection to semi-space algorithm
Posted by Xiao-Feng Li <xi...@gmail.com>.
On Fri, Mar 21, 2008 at 5:02 PM, Stepan Mishura
<st...@gmail.com> wrote:
> On 3/20/08, Xiao-Feng Li <xi...@gmail.com> wrote:
> > Hello, since the semi-space algortithm recently developed in Harmony
> > GC has obviously better performance than the current partial-forward
> > algorithm when they are used for minor collection, I plan to switch
> > the default minor collection to semi-space algorithm. The results in
> > my runs have no new failures with classlib unit tests, drlvm tests and
> > reg tests, compared to current default GC on Linux/Win 32/64-bit
> > platforms. Although I have strong confidence that the failure cases
> > should be minimum, I still expect there be some regressions with other
> > untested test cases. I hope the semispace GC to be the default in next
> > milestone build. Please forgive me if the switch causes some
> > unexpected failures, and GC people will try best to fix the failures
> > ASAP.
> >
>
> Hi Xiaofeng,
>
> The current snapshot under testing is r639357. As I see you turned on
> semispace algorithm at r639530. The difference between r639357 and
> r639530 is only your update: "turn on semispace algorithm as the
> default minor collection GC"
>
> So we can build and test r639530 snapshot and compare with results for
> r639357. That may help you to detect regressions caused by the update.
Great, this helps!
Thanks,
xiaofeng
> Thanks,
> Stepan.
>
>
>
> > Please let me know if anybody has objections. If no objections, I will
> > make the switch in 12 hours. Thanks for your supports.
> >
> > Thanks,
> > xiaofeng
> >
> > --
> > http://xiao-feng.blogspot.com
> >
>
--
http://xiao-feng.blogspot.com
Re: [drlvm][gc] switch the default GC minor collection to semi-space algorithm
Posted by Stepan Mishura <st...@gmail.com>.
On 3/20/08, Xiao-Feng Li <xi...@gmail.com> wrote:
> Hello, since the semi-space algortithm recently developed in Harmony
> GC has obviously better performance than the current partial-forward
> algorithm when they are used for minor collection, I plan to switch
> the default minor collection to semi-space algorithm. The results in
> my runs have no new failures with classlib unit tests, drlvm tests and
> reg tests, compared to current default GC on Linux/Win 32/64-bit
> platforms. Although I have strong confidence that the failure cases
> should be minimum, I still expect there be some regressions with other
> untested test cases. I hope the semispace GC to be the default in next
> milestone build. Please forgive me if the switch causes some
> unexpected failures, and GC people will try best to fix the failures
> ASAP.
>
Hi Xiaofeng,
The current snapshot under testing is r639357. As I see you turned on
semispace algorithm at r639530. The difference between r639357 and
r639530 is only your update: "turn on semispace algorithm as the
default minor collection GC"
So we can build and test r639530 snapshot and compare with results for
r639357. That may help you to detect regressions caused by the update.
Thanks,
Stepan.
> Please let me know if anybody has objections. If no objections, I will
> make the switch in 12 hours. Thanks for your supports.
>
> Thanks,
> xiaofeng
>
> --
> http://xiao-feng.blogspot.com
>