You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@harmony.apache.org by Robin Garner <ro...@anu.edu.au> on 2007/11/15 08:43:58 UTC
Re: [drlvm]Leave classunloading on by default?
Rana Dasgupta wrote:
> BTW, I have been unable to find a scenario that stresses the gc
> heap, and needs to unload unused classes, and it would be good find
> such test cases.
The DaCapo benchmarks project is currently evaluating new benchmarks,
and in svn right now there is a derby benchmark based on JDBCBench (ie
almost a direct copy of the hsqldb benchmark).
Because derby compiles SQL queries into Java classes, after a couple of
iterations of the benchmark, it requires class unloading. The benchmark
appears to run in a ~25M heap, and allocates much more than that.
It's almost certain that we won't release this as the final derby
benchmark, but I might package it separately as a class-unloading unit
test. But in the meantime, if you want to build dacapo from svn head,
it might be exactly what you're after.
cheers
--
Robin Garner
Dept. of Computer Science
Australian National University
http://cs.anu.edu.au/people/Robin.Garner/
Re: [drlvm] Class unloading benchmark (was: Re: [drlvm]Leave classunloading on by default?)
Posted by Rana Dasgupta <rd...@gmail.com>.
Thanks will try this out.
Rana
On Jan 17, 2008 12:14 AM, Robin Garner <ro...@anu.edu.au> wrote:
> Rana Dasgupta wrote:
> > Thanks :-)
> >
> > On Nov 21, 2007 6:28 PM, Robin Garner <ro...@anu.edu.au> wrote:
> >
> >
> >> I'm glad it's of use. It shouldn't be too hard to make a standalone
> >> test from it, I'll do that next week sometime.
> >>
> >> cheers
> >>
> >> Rana Dasgupta wrote:
> >>
> >>> Robin,
> >>> I built dacapo from svn head and ran the derby benchmark. I see a
> few
> >>> failures in DRLVM in class unloading( when using mark&compact ),
> >>>
> >> shutdown
> >>
> >>> etc. when running with different heap sizes. The RI successfully runs
> >>> everything. As I said before, a real use case always helps. Are you
> >>>
> >> planning
> >>
> >>> to contribute this as a standalone test? That would help a lot. I
> don't
> >>> want to open up JIRA's that need the entire dacapo svn.
> >>>
> >>> Thanks,
> >>> Rana
> >>>
> The attached tarball is a standalone version of the derby class
> unloading benchmark. Unpack it, run 'ant' in the directory it creates,
> then run 'runme.sh'. There are some options to play with.
>
> cheers
> >>> On Nov 15, 2007 12:43 AM, Robin Garner <ro...@anu.edu.au>
> wrote:
> >>>
> >>>
> >>>> Rana Dasgupta wrote:
> >>>>
> >>>>> BTW, I have been unable to find a scenario that stresses the gc
> >>>>> heap, and needs to unload unused classes, and it would be good find
> >>>>> such test cases.
> >>>>>
> >>>> The DaCapo benchmarks project is currently evaluating new benchmarks,
> >>>> and in svn right now there is a derby benchmark based on JDBCBench
> (ie
> >>>> almost a direct copy of the hsqldb benchmark).
> >>>>
> >>>> Because derby compiles SQL queries into Java classes, after a couple
> of
> >>>> iterations of the benchmark, it requires class unloading. The
> >>>>
> >> benchmark
> >>
> >>>> appears to run in a ~25M heap, and allocates much more than that.
> >>>>
> >>>> It's almost certain that we won't release this as the final derby
> >>>> benchmark, but I might package it separately as a class-unloading
> unit
> >>>> test. But in the meantime, if you want to build dacapo from svn
> head,
> >>>> it might be exactly what you're after.
> >>>>
> >>>> cheers
> >>>>
> >>>> --
> >>>> Robin Garner
> >>>> Dept. of Computer Science
> >>>> Australian National University
> >>>> http://cs.anu.edu.au/people/Robin.Garner/
> >>>>
> >>>>
> >> --
> >> Robin Garner
> >> Dept. of Computer Science
> >> Australian National University
> >> http://cs.anu.edu.au/people/Robin.Garner/
> >>
> >>
> >
> >
>
Re: [drlvm] Class unloading benchmark (was: Re: [drlvm]Leave classunloading on by default?)
Posted by Xiao-Feng Li <xi...@gmail.com>.
On Jan 17, 2008 3:14 PM, Robin Garner <ro...@anu.edu.au> wrote:
> Rana Dasgupta wrote:
> > Thanks :-)
> >
> > On Nov 21, 2007 6:28 PM, Robin Garner <ro...@anu.edu.au> wrote:
> >
> >
> >> I'm glad it's of use. It shouldn't be too hard to make a standalone
> >> test from it, I'll do that next week sometime.
> >>
> >> cheers
> >>
> >> Rana Dasgupta wrote:
> >>
> >>> Robin,
> >>> I built dacapo from svn head and ran the derby benchmark. I see a few
> >>> failures in DRLVM in class unloading( when using mark&compact ),
> >>>
> >> shutdown
> >>
> >>> etc. when running with different heap sizes. The RI successfully runs
> >>> everything. As I said before, a real use case always helps. Are you
> >>>
> >> planning
> >>
> >>> to contribute this as a standalone test? That would help a lot. I don't
> >>> want to open up JIRA's that need the entire dacapo svn.
> >>>
> >>> Thanks,
> >>> Rana
> >>>
> The attached tarball is a standalone version of the derby class
> unloading benchmark. Unpack it, run 'ant' in the directory it creates,
> then run 'runme.sh'. There are some options to play with.
Thanks! Will try it.
-xiaofeng
> cheers
> >>> On Nov 15, 2007 12:43 AM, Robin Garner <ro...@anu.edu.au> wrote:
> >>>
> >>>
> >>>> Rana Dasgupta wrote:
> >>>>
> >>>>> BTW, I have been unable to find a scenario that stresses the gc
> >>>>> heap, and needs to unload unused classes, and it would be good find
> >>>>> such test cases.
> >>>>>
> >>>> The DaCapo benchmarks project is currently evaluating new benchmarks,
> >>>> and in svn right now there is a derby benchmark based on JDBCBench (ie
> >>>> almost a direct copy of the hsqldb benchmark).
> >>>>
> >>>> Because derby compiles SQL queries into Java classes, after a couple of
> >>>> iterations of the benchmark, it requires class unloading. The
> >>>>
> >> benchmark
> >>
> >>>> appears to run in a ~25M heap, and allocates much more than that.
> >>>>
> >>>> It's almost certain that we won't release this as the final derby
> >>>> benchmark, but I might package it separately as a class-unloading unit
> >>>> test. But in the meantime, if you want to build dacapo from svn head,
> >>>> it might be exactly what you're after.
> >>>>
> >>>> cheers
> >>>>
> >>>> --
> >>>> Robin Garner
> >>>> Dept. of Computer Science
> >>>> Australian National University
> >>>> http://cs.anu.edu.au/people/Robin.Garner/
> >>>>
> >>>>
> >> --
> >> Robin Garner
> >> Dept. of Computer Science
> >> Australian National University
> >> http://cs.anu.edu.au/people/Robin.Garner/
> >>
> >>
> >
> >
>
--
http://xiao-feng.blogspot.com
[drlvm] Class unloading benchmark (was: Re: [drlvm]Leave classunloading
on by default?)
Posted by Robin Garner <ro...@anu.edu.au>.
Rana Dasgupta wrote:
> Thanks :-)
>
> On Nov 21, 2007 6:28 PM, Robin Garner <ro...@anu.edu.au> wrote:
>
>
>> I'm glad it's of use. It shouldn't be too hard to make a standalone
>> test from it, I'll do that next week sometime.
>>
>> cheers
>>
>> Rana Dasgupta wrote:
>>
>>> Robin,
>>> I built dacapo from svn head and ran the derby benchmark. I see a few
>>> failures in DRLVM in class unloading( when using mark&compact ),
>>>
>> shutdown
>>
>>> etc. when running with different heap sizes. The RI successfully runs
>>> everything. As I said before, a real use case always helps. Are you
>>>
>> planning
>>
>>> to contribute this as a standalone test? That would help a lot. I don't
>>> want to open up JIRA's that need the entire dacapo svn.
>>>
>>> Thanks,
>>> Rana
>>>
The attached tarball is a standalone version of the derby class
unloading benchmark. Unpack it, run 'ant' in the directory it creates,
then run 'runme.sh'. There are some options to play with.
cheers
>>> On Nov 15, 2007 12:43 AM, Robin Garner <ro...@anu.edu.au> wrote:
>>>
>>>
>>>> Rana Dasgupta wrote:
>>>>
>>>>> BTW, I have been unable to find a scenario that stresses the gc
>>>>> heap, and needs to unload unused classes, and it would be good find
>>>>> such test cases.
>>>>>
>>>> The DaCapo benchmarks project is currently evaluating new benchmarks,
>>>> and in svn right now there is a derby benchmark based on JDBCBench (ie
>>>> almost a direct copy of the hsqldb benchmark).
>>>>
>>>> Because derby compiles SQL queries into Java classes, after a couple of
>>>> iterations of the benchmark, it requires class unloading. The
>>>>
>> benchmark
>>
>>>> appears to run in a ~25M heap, and allocates much more than that.
>>>>
>>>> It's almost certain that we won't release this as the final derby
>>>> benchmark, but I might package it separately as a class-unloading unit
>>>> test. But in the meantime, if you want to build dacapo from svn head,
>>>> it might be exactly what you're after.
>>>>
>>>> cheers
>>>>
>>>> --
>>>> Robin Garner
>>>> Dept. of Computer Science
>>>> Australian National University
>>>> http://cs.anu.edu.au/people/Robin.Garner/
>>>>
>>>>
>> --
>> Robin Garner
>> Dept. of Computer Science
>> Australian National University
>> http://cs.anu.edu.au/people/Robin.Garner/
>>
>>
>
>
Re: [drlvm]Leave classunloading on by default?
Posted by Rana Dasgupta <rd...@gmail.com>.
Thanks :-)
On Nov 21, 2007 6:28 PM, Robin Garner <ro...@anu.edu.au> wrote:
> I'm glad it's of use. It shouldn't be too hard to make a standalone
> test from it, I'll do that next week sometime.
>
> cheers
>
> Rana Dasgupta wrote:
> > Robin,
> > I built dacapo from svn head and ran the derby benchmark. I see a few
> > failures in DRLVM in class unloading( when using mark&compact ),
> shutdown
> > etc. when running with different heap sizes. The RI successfully runs
> > everything. As I said before, a real use case always helps. Are you
> planning
> > to contribute this as a standalone test? That would help a lot. I don't
> > want to open up JIRA's that need the entire dacapo svn.
> >
> > Thanks,
> > Rana
> >
> > On Nov 15, 2007 12:43 AM, Robin Garner <ro...@anu.edu.au> wrote:
> >
> >> Rana Dasgupta wrote:
> >>> BTW, I have been unable to find a scenario that stresses the gc
> >>> heap, and needs to unload unused classes, and it would be good find
> >>> such test cases.
> >> The DaCapo benchmarks project is currently evaluating new benchmarks,
> >> and in svn right now there is a derby benchmark based on JDBCBench (ie
> >> almost a direct copy of the hsqldb benchmark).
> >>
> >> Because derby compiles SQL queries into Java classes, after a couple of
> >> iterations of the benchmark, it requires class unloading. The
> benchmark
> >> appears to run in a ~25M heap, and allocates much more than that.
> >>
> >> It's almost certain that we won't release this as the final derby
> >> benchmark, but I might package it separately as a class-unloading unit
> >> test. But in the meantime, if you want to build dacapo from svn head,
> >> it might be exactly what you're after.
> >>
> >> cheers
> >>
> >> --
> >> Robin Garner
> >> Dept. of Computer Science
> >> Australian National University
> >> http://cs.anu.edu.au/people/Robin.Garner/
> >>
> >
>
>
> --
> Robin Garner
> Dept. of Computer Science
> Australian National University
> http://cs.anu.edu.au/people/Robin.Garner/
>
Re: [drlvm]Leave classunloading on by default?
Posted by Robin Garner <ro...@anu.edu.au>.
I'm glad it's of use. It shouldn't be too hard to make a standalone
test from it, I'll do that next week sometime.
cheers
Rana Dasgupta wrote:
> Robin,
> I built dacapo from svn head and ran the derby benchmark. I see a few
> failures in DRLVM in class unloading( when using mark&compact ), shutdown
> etc. when running with different heap sizes. The RI successfully runs
> everything. As I said before, a real use case always helps. Are you planning
> to contribute this as a standalone test? That would help a lot. I don't
> want to open up JIRA's that need the entire dacapo svn.
>
> Thanks,
> Rana
>
> On Nov 15, 2007 12:43 AM, Robin Garner <ro...@anu.edu.au> wrote:
>
>> Rana Dasgupta wrote:
>>> BTW, I have been unable to find a scenario that stresses the gc
>>> heap, and needs to unload unused classes, and it would be good find
>>> such test cases.
>> The DaCapo benchmarks project is currently evaluating new benchmarks,
>> and in svn right now there is a derby benchmark based on JDBCBench (ie
>> almost a direct copy of the hsqldb benchmark).
>>
>> Because derby compiles SQL queries into Java classes, after a couple of
>> iterations of the benchmark, it requires class unloading. The benchmark
>> appears to run in a ~25M heap, and allocates much more than that.
>>
>> It's almost certain that we won't release this as the final derby
>> benchmark, but I might package it separately as a class-unloading unit
>> test. But in the meantime, if you want to build dacapo from svn head,
>> it might be exactly what you're after.
>>
>> cheers
>>
>> --
>> Robin Garner
>> Dept. of Computer Science
>> Australian National University
>> http://cs.anu.edu.au/people/Robin.Garner/
>>
>
--
Robin Garner
Dept. of Computer Science
Australian National University
http://cs.anu.edu.au/people/Robin.Garner/
Re: [drlvm]Leave classunloading on by default?
Posted by Rana Dasgupta <rd...@gmail.com>.
Robin,
I built dacapo from svn head and ran the derby benchmark. I see a few
failures in DRLVM in class unloading( when using mark&compact ), shutdown
etc. when running with different heap sizes. The RI successfully runs
everything. As I said before, a real use case always helps. Are you planning
to contribute this as a standalone test? That would help a lot. I don't
want to open up JIRA's that need the entire dacapo svn.
Thanks,
Rana
On Nov 15, 2007 12:43 AM, Robin Garner <ro...@anu.edu.au> wrote:
> Rana Dasgupta wrote:
> > BTW, I have been unable to find a scenario that stresses the gc
> > heap, and needs to unload unused classes, and it would be good find
> > such test cases.
>
> The DaCapo benchmarks project is currently evaluating new benchmarks,
> and in svn right now there is a derby benchmark based on JDBCBench (ie
> almost a direct copy of the hsqldb benchmark).
>
> Because derby compiles SQL queries into Java classes, after a couple of
> iterations of the benchmark, it requires class unloading. The benchmark
> appears to run in a ~25M heap, and allocates much more than that.
>
> It's almost certain that we won't release this as the final derby
> benchmark, but I might package it separately as a class-unloading unit
> test. But in the meantime, if you want to build dacapo from svn head,
> it might be exactly what you're after.
>
> cheers
>
> --
> Robin Garner
> Dept. of Computer Science
> Australian National University
> http://cs.anu.edu.au/people/Robin.Garner/
>
Re: [drlvm]Leave classunloading on by default?
Posted by Alexei Fedotov <al...@gmail.com>.
> it may inspire somebody to write
> a stress test that creates lots of new classes in order to exercise
> Harmony's class unloading code.
Four stress tests on class unloading were written by Nikolay Chugunov
last year. They are currently in the exclude list of the stress test
suite. I wonder if they should be enabled (= class unloading is on by
default).
On 11/17/07, Tim Ellison <t....@gmail.com> wrote:
> Robin Garner wrote:
> > Rana Dasgupta wrote:
> >> BTW, I have been unable to find a scenario that stresses the gc
> >> heap, and needs to unload unused classes, and it would be good find
> >> such test cases.
> >
> > The DaCapo benchmarks project is currently evaluating new benchmarks,
> > and in svn right now there is a derby benchmark based on JDBCBench (ie
> > almost a direct copy of the hsqldb benchmark).
> >
> > Because derby compiles SQL queries into Java classes, after a couple of
> > iterations of the benchmark, it requires class unloading. The benchmark
> > appears to run in a ~25M heap, and allocates much more than that.
> >
> > It's almost certain that we won't release this as the final derby
> > benchmark, but I might package it separately as a class-unloading unit
> > test. But in the meantime, if you want to build dacapo from svn head,
> > it might be exactly what you're after.
>
> ack -- thanks Robin. At the very least it may inspire somebody to write
> a stress test that creates lots of new classes in order to exercise
> Harmony's class unloading code.
>
> Regards,
> Tim
>
--
With best regards,
Alexei,
ESSD, Intel
Re: [drlvm]Leave classunloading on by default?
Posted by Rana Dasgupta <rd...@gmail.com>.
It sounds like a more realistic test than the ones we been using (thanks
Robin ). I am tied up a little , but yes I intend to try it.
Rana
On 19 Nov 2007 10:55:06 +0300, Egor Pasko <eg...@gmail.com> wrote:
> On the 0x393 day of Apache Harmony Rana Dasgupta wrote:
> > Could someone please commit JIRA 4931 to turn on the feature by default?
>
> did you check it on DaCapo from svn? just interesting..
>
> > Thanks,
> > Rana
> >
> > On Nov 16, 2007 2:57 PM, Tim Ellison <t....@gmail.com> wrote:
> >
> > > Robin Garner wrote:
> > > > Rana Dasgupta wrote:
> > > >> BTW, I have been unable to find a scenario that stresses the gc
> > > >> heap, and needs to unload unused classes, and it would be good find
> > > >> such test cases.
> > > >
> > > > The DaCapo benchmarks project is currently evaluating new
> benchmarks,
> > > > and in svn right now there is a derby benchmark based on JDBCBench
> (ie
> > > > almost a direct copy of the hsqldb benchmark).
> > > >
> > > > Because derby compiles SQL queries into Java classes, after a couple
> of
> > > > iterations of the benchmark, it requires class unloading. The
> benchmark
> > > > appears to run in a ~25M heap, and allocates much more than that.
> > > >
> > > > It's almost certain that we won't release this as the final derby
> > > > benchmark, but I might package it separately as a class-unloading
> unit
> > > > test. But in the meantime, if you want to build dacapo from svn
> head,
> > > > it might be exactly what you're after.
> > >
> > > ack -- thanks Robin. At the very least it may inspire somebody to
> write
> > > a stress test that creates lots of new classes in order to exercise
> > > Harmony's class unloading code.
> > >
> > > Regards,
> > > Tim
> > >
>
> --
> Egor Pasko
>
>
Re: [drlvm]Leave classunloading on by default?
Posted by Egor Pasko <eg...@gmail.com>.
On the 0x393 day of Apache Harmony Rana Dasgupta wrote:
> Could someone please commit JIRA 4931 to turn on the feature by default?
did you check it on DaCapo from svn? just interesting..
> Thanks,
> Rana
>
> On Nov 16, 2007 2:57 PM, Tim Ellison <t....@gmail.com> wrote:
>
> > Robin Garner wrote:
> > > Rana Dasgupta wrote:
> > >> BTW, I have been unable to find a scenario that stresses the gc
> > >> heap, and needs to unload unused classes, and it would be good find
> > >> such test cases.
> > >
> > > The DaCapo benchmarks project is currently evaluating new benchmarks,
> > > and in svn right now there is a derby benchmark based on JDBCBench (ie
> > > almost a direct copy of the hsqldb benchmark).
> > >
> > > Because derby compiles SQL queries into Java classes, after a couple of
> > > iterations of the benchmark, it requires class unloading. The benchmark
> > > appears to run in a ~25M heap, and allocates much more than that.
> > >
> > > It's almost certain that we won't release this as the final derby
> > > benchmark, but I might package it separately as a class-unloading unit
> > > test. But in the meantime, if you want to build dacapo from svn head,
> > > it might be exactly what you're after.
> >
> > ack -- thanks Robin. At the very least it may inspire somebody to write
> > a stress test that creates lots of new classes in order to exercise
> > Harmony's class unloading code.
> >
> > Regards,
> > Tim
> >
--
Egor Pasko
Re: [drlvm]Leave classunloading on by default?
Posted by Rana Dasgupta <rd...@gmail.com>.
Could someone please commit JIRA 4931 to turn on the feature by default?
Thanks,
Rana
On Nov 16, 2007 2:57 PM, Tim Ellison <t....@gmail.com> wrote:
> Robin Garner wrote:
> > Rana Dasgupta wrote:
> >> BTW, I have been unable to find a scenario that stresses the gc
> >> heap, and needs to unload unused classes, and it would be good find
> >> such test cases.
> >
> > The DaCapo benchmarks project is currently evaluating new benchmarks,
> > and in svn right now there is a derby benchmark based on JDBCBench (ie
> > almost a direct copy of the hsqldb benchmark).
> >
> > Because derby compiles SQL queries into Java classes, after a couple of
> > iterations of the benchmark, it requires class unloading. The benchmark
> > appears to run in a ~25M heap, and allocates much more than that.
> >
> > It's almost certain that we won't release this as the final derby
> > benchmark, but I might package it separately as a class-unloading unit
> > test. But in the meantime, if you want to build dacapo from svn head,
> > it might be exactly what you're after.
>
> ack -- thanks Robin. At the very least it may inspire somebody to write
> a stress test that creates lots of new classes in order to exercise
> Harmony's class unloading code.
>
> Regards,
> Tim
>
Re: [drlvm]Leave classunloading on by default?
Posted by Tim Ellison <t....@gmail.com>.
Robin Garner wrote:
> Rana Dasgupta wrote:
>> BTW, I have been unable to find a scenario that stresses the gc
>> heap, and needs to unload unused classes, and it would be good find
>> such test cases.
>
> The DaCapo benchmarks project is currently evaluating new benchmarks,
> and in svn right now there is a derby benchmark based on JDBCBench (ie
> almost a direct copy of the hsqldb benchmark).
>
> Because derby compiles SQL queries into Java classes, after a couple of
> iterations of the benchmark, it requires class unloading. The benchmark
> appears to run in a ~25M heap, and allocates much more than that.
>
> It's almost certain that we won't release this as the final derby
> benchmark, but I might package it separately as a class-unloading unit
> test. But in the meantime, if you want to build dacapo from svn head,
> it might be exactly what you're after.
ack -- thanks Robin. At the very least it may inspire somebody to write
a stress test that creates lots of new classes in order to exercise
Harmony's class unloading code.
Regards,
Tim