[ptt-users] JVM parameters to tune TestMaker on Solaris

Fri Mar 2 04:33:29 PST 2007

Hi Kevin,

I could test your requirements during my last load-test project, and results
were very  good : injectors were steady during all my tests.

Thx a lot !

Pierre.

2006/9/17, Dyer, Kevin <kevin.dyer at matrixone.com>:
>
>  Salut Pierre,
>
> As I reread the description I also noticed that, at least in 1.4 JVM test
> case, you didn't specify the SurvivorRatio. With our product testing a
> SurvivorRatio set to 3 improve the Eden space usage.
>
> Hope this helps.
>
>
> Kevin
>
> --------------------------
> Kevin J. Dyer| Knowledge Management and Customer Advocacy, Technical
> Support
>  | ENOVIA MatrixOne
> 210 Littleton Road | Westford, MA  01886
> Office: +1 978 589 4165 | Fax: +1 978 589 5700 | Cell: +1 978 549 0971
> kevin.dyer at MatrixOne.com
>
>
> -----Original Message-----
> From: users-bounces at lists.pushtotest.com <
> users-bounces at lists.pushtotest.com>
> To: TestMaker users list <users at lists.pushtotest.com>
> Sent: Sat Sep 16 08:23:53 2006
> Subject: Re: [ptt-users] JVM parameters to tune TestMaker on Solaris
>
> Hi,
>
> I worked to issue this subject. I would like to work again but I have
> no time any more.
>
> So, I report back my current observations and I hope that could help some
> one...
>
> I worked with 3 OS and i used an "Hello World" scenario (2000VU -
> 4VU/s - 30mn) :
>  - a Intel Desktop (XP - 1Go Ram)
>  - a Sun Fire 280R (SunOS9 - 4Go)
>  - a Bipro Intel (Lunix RH4 - 4 Go Ram)
> Hello World tests generates a heavy activity in the Eden part... so,
> configurations were oriented to optimize the GC behaviour, in these
> conditions.
>
> I was surpised to have the best throughput with my poor desktop... but
> XP is limited in socket allocations.
> On the Solaris, I bench JVM 1.4 and JVM 1.5... and JVM 1.5 behaviour is
> better.
> On the Linux, I validate that BEA Jrockit is supported, but I have no
> time to compare with Sun JVM.
>
>
> Now, my best configurations :
>
> - SUN SOLARIS JVM 1.5 :
>    . Mode server
>    . Xmx 2048m
>    . Xms 1024m
>    . Xss 256k
>    . XX:MaxNewSize 256m
>    . XX:NewSize 256m
>    . XX:MaxPermSize 192m
>    . UseParalleleGC (CMSGC Incr is good too)
>    . ParallelGCThread 8
>    . GCTimeRatio 999 (only supported by JVM 1.5)
>    . XX:NewRatio 1 (only supported by JVM 1.5)
> With this configuration, Injector applied 1960 transaction/s against
> the theoretical 2000 Transaction/s . The GC Time was stable roundabout
> 100ms... once in 8s.
>
> - WINDOWS XP - JVM 1.4
>    . Mode server
>    . Xmx 512m
>    . Xms 256m
>    . Xss 128k
>    . XX:MaxNewSize 128m
>    . XX:NewSize 128m
>    . XX:MaxPermSize 96m
>    . UseParalleleGC
> With this configuration, Injector applied 1970 transaction/s against
> the theoretical 2000 Transaction/s . The GC Time was stable roundabout
> 70ms... once in 5s.
>
> - Linux - BEA Jrockit 1.4
>    . Mode server
>    . Xmx 512m
>    . Xms 256m
>    . Xss 128k
>    . Xgc:parallel
>    . Xgcprio:throughput
>    . Xcleartype:local
>  With this configuration, Injector applied 1870 transaction/s against
> the theoretical 2000 Transaction/s . The GC Time was stable roundabout
> 220ms... once in 30s.
>
> "Hello world scenario" is not a representative test, but it is the
> main to discover the black art... and I hope i could start back from
> these observations when i have time to work on these subjects.
>
> Pierre Laboureau.
>
>
> 2005/10/19, Frank Cohen <fcohen at pushtotest.com>:
> > Hi Pierre: I am reminded of Star Wars where Annikan asks Palpatine
> > "Can I learn these skills?" and Palpatine answers "Not from a Jedi."
> > Tuning Java on Solaris is a black art. Results depend greatly on
> > which version of Java you are using. Most of my experience is on JVM
> > 1.4.2 on Solaris 9. So I'm not the Sith to help you.
> >
> > May I recommend you take a look at the optimization settings we made
> > for the SOA Performance Kit project. The Developer Journal has a
> > little section on it. See http://www.pushtotest.com/Downloads/kits/
> > soakit.html
> >
> > I started a new wiki page for tuning TestMaker tips at:
> > http://www.pushtotest.com/Wiki/TestMakerTuning
> >
> > By the way, as far as I know the Jython interpreter that TestMaker
> > embeds is not fully compatible with JVM 1.5. It uses the enum key
> > word. I am participating in the Jython developer group and expect a
> > new Jython - with many new features and 1.5 compatibility - will be
> > in TestMaker by the end of this year.
> >
> > Lastly, another way to go is to use the TestMaker Platform. This is a
> > commercial version of TestMaker that implements the TestNetwork
> > distributed environment to run scripts on multiple TestNodes. Details
> > are found here:
> > http://www.pushtotest.com/Downloads/features.html
> >
> > Hope this helps.
> >
> > -Frank
> >
> >
> >
> > On Oct 18, 2005, at 11:55 PM, Pierre Laboureau wrote:
> >
> > > Hi,
> > >
> > > I am working on load tests with TestMaker (XsTest). I am trying to
> > > generate "Hello World" with 1000 VU. Around 400 VU, I observe some
> > > problems about performance :
> > > - Out Of Memory
> > > - Increase of GC time
> > > - Long response times (Threads bottleneck ?)
> > > - ....
> > >
> > > I am looking for tune the JVM (1.5.02)... but there are lots of
> > > options, and more combinations to tune JVM HotSpot :
> > > Compilator : -server, -client
> > > Memory : -Xmx, -Xms, Xss
> > > Thread synchronisation : -XX:-UseLWPSynchronization (Many-to-many
> > > model), -XX:+UseBoundThreads (One-to-One model)... -Xconcurrentio
> > > (obsolete ?)
> > > Garbage Collector : -XX:+UseSerialGC, -XX:+UseParallelGC,
> > > -XX:GCTimeRatio,  -XX:ParallelGCThreads....
> > >
> > > I am still working on  ! But, is there a base document which
> > > recommends JVM parameters using TestMaker ? With Solaris 9 ? With
> > > other OS ? With a JVM 1.4 ? With a JVM 1.5 ?
> > > _______________________________________________
> > > Users mailing list
> > > Users at lists.pushtotest.com
> > > http://lists.pushtotest.com/mailman/listinfo/users
> > >
> > >
> >
> > _______________________________________________
> > Users mailing list
> > Users at lists.pushtotest.com
> > http://lists.pushtotest.com/mailman/listinfo/users
> >
> _______________________________________________
> Users mailing list
> Users at lists.pushtotest.com
> http://lists.pushtotest.com/mailman/listinfo/users
>
>
> _______________________________________________
> Users mailing list
> Users at lists.pushtotest.com
> http://lists.pushtotest.com/mailman/listinfo/users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cake.pushtotest.com/pipermail/users/attachments/20070302/b39900f1/attachment-0001.htm