Re: [Java-gnome-developer] JVM benchmarks: kaffe takes the cake

[Clemens E. <lin...@we...>]

Hi Tiago!

Your benchmarks ment nice, but they are not really useful.

1.) What does the *9, *3 mean for memory-consumption?
Have you ever heard of threads? Threads are under Linux(o.k., thats not 
really correct) processes that share stuff between. That means Sun´s JVM 
does need 7,5% of your memory, not 67,5%.

2.) Your startup-results seem that you´ve started the JVMs already, not 
really fair, because JVMs like Sun´s one need to load a lot of shared 
stuff. It would be fair to measure with and without haveing started it 
before and listing both values.

3.) You forgot a really important point: Execution speed.
GIJ wasnt designed to be fast, it runs java-code only in interpreted 
mode, because it was ment to be a way for GCJ to load dynamically 
classes. Kaffe has two fifferent JITs.
IBMs JVM is the fastest one, when only measuring execution speed. -> 100%
After IBM theres Sun. -> 90%
after Sun theres kaffe ->25-45%
after kaffe theres GIJ -> 5-10%

However these percentages are not 100% correct and depend heavily on 
what you do with the VMs.

However IBMs VM is based on Suns classes, that means Kaffe is the 
fastest OPenSource-JVM laying arround. Suns&IBMs JVMs are as fast or 
faster than GCJ, which compiles java statically. Even GCJ developers 
commented that Sun-JVM is impressive!

4.)
 > better choice then Sun's java. This makes you wonder how good is Sun
 > investing in the Linux platform.
Sun is currently rewriting their whole Java2D/x11 rendering pipeline.
They will use OpenGL starting from 1.5 for drawing things. I´ve spoken 
with a Java2D-enginier and he told me that the results are impressive.
If Linux would not matter, they wouldnt spend so much time improving it 
on Linux/Desktop.

lg Clemens