Java Virtual Machine Benchmarks

[savageautomate]

Java VM benchmarks for the various JVMs available for Raspberry Pi:
http://www.savagehomeautomation.com/pro ... marks.html

[xranby]

Thank you for starting this thread, Dacapo is a nice benchamrk that gives a honest view how fast a JVM can solve large batch jobs.

I have done my own benchmarks using Scimark 2
The Scimark 2 benchmark is good to demonstrate CPU loop optimizations.

The raw numbers:

Relative performance compared to the zero interpreted-mode:


For best results:
Use ahead of time compilation using GCJ on the Pi running armhf.
For good results try:
Oracle JDK 7u6 on armel.
CACAO gives good results for armel and openjdk-6.
Avian also gives good results for armhf and openjdk-7.
JamVM gives ok results for armhf and openjdk-6/7.

The monte carlo benchmark is interesting, here the Oracle JDK clearly got an advantage by being 4 times faster compared to all FOSS options.

References:
How to use GCJ: http://www.raspberrypi.org/phpBB3/viewt ... 79#p159779
How to insall Oracle JDK 7u6 for armel : http://www.savagehomeautomation.com/pro ... -170u.html
How to build avian + OpenJDK 7 for armhf : http://labb.zafena.se/?p=630
How to build JamVM + OpenJDK 6/7 for armhf : http://labb.zafena.se/?p=576


I am still looking for a good benchmark to measure responsiveness in client applications.
it would be interesting to check which JVM would be most suitable for audio and 3D game programming on the Pi.

Cheers
Xerxes

[xranby]

Yesterday i experimented running a real-time java arcade game on the Pi written using Java2D and deployed using javaws JNLP.
http://www.raspberrypi.org/phpBB3/viewt ... 14#p177614
Here JamVM came out as a clear winner on the Pi for realtime java game applications:
OpenJDK 7 + JamVM + icedtea-web gave 28 fps in game using Rasbian armhf
OpenJDK 7 + zero mixed mode + icedtea-web gave 21fps in game fps using Rasbian armhf
Oracle 7u6 JDK using Debian armel was unable to start the game since it did not contain a javaws JNLP launcher.

This clearly indicate that we need to have better benchmarks for realtime java use cases like interactive 2D applications.