--- On Sat, 9/13/08, Mukul Gandhi <[EMAIL PROTECTED]> wrote:

...
> Also I wrote, the execution timing on both the OSs is < 1 second. To
> my opinion, you can ignore such a small difference.

Or more precisely: the methodology for testing seems flawed. Because of the way 
modern JVMs work, single run without sufficient warmup time and repetitions 
only tests how fast JVM can ramp up things.
It tells very little about application or library performance.

To get a realistic picture, one needs to run multiple transformations; first to 
"warm up" JVM (since it will load in system libraries, hot spot compiler will 
compile some byte code to native code etc), and only then measure. And measure 
multiple runs as well.

-+ Tatu +-



      

Reply via email to