Re: [j-devel] ABCL performance on Java 1.6SE

[Mark E. <ev...@pa...>]

Some numbers for Java 5 vs. Java 6 performance. . .

I picked recompiling its own fasls as a starting point for testing ABCL 
performance differences.


 From http://code.google.com/p/abcl-dynamic-install/wiki/Performance


  Platform 	  JDK                   Time (Seconds) 	
x86-solaris-5.11  sun-jdk-1.5.0_16 	144 177 144.561 	
x86-solaris-5.11  sun-jdk-1.6.0_07 	248 315 258.914 272.562 245.975	
x86-solaris-5.11  sun-jdk-1.6.0_10rc 	153.252 143.249 141.235	
x86-freebsd-7.0   bsd-jdk-1.5.0_15-p8 	128.182 128.345 127.595 	
x86-freebsd-7.0   bsd-jdk-1.6.0_03-p4 	236.737 224.343 232.728	
x64-darwin-9.4.0  apple-jdk-1.6.0_05 	117.68 171.89 141.289 	
x64-darwin-9.4.0  apple-jdk-1.5.0_13 	104.683 106.769 	


Trials with the same 'Platform' were performed on the same physical 
machine.

So, Java 6 performance for this task in generally worse than Java 5 on 
the same platform, with "newer" versions of Java 6 doing increasingly 
better.  Interestingly, by the JDK-1.6.0_10rc release on x86-solaris, 
Java 6 has roughly "caught" up with the best JDK-1.5.0 performance on 
that platform.

I would conclude that if we choose to target "performance on Java 5" for 
ABCL 1.0, "performance on Java 6" will also be met.

The actual profiling tools under Netbeans 6.1 seem pretty good, but 
again I really don't have a strategy yet for profiling ABCL.

Mark


-- 
"A screaming comes across the sky.  It has happened before, but there is
nothing to compare to it now."