Posted by kirillcool
on December 13, 2005 at 8:24 AM PST
After numerous press releases showing that BEA's JRockit is the fastest JVM around, I have decided to try it out for a desktop application. The result - it may be the fastest for the application servers, but it most certainly is not the best for the desktop.
In the world of seemingly arbitrary records set by JRockit 5.0 JVM from BEA (see this comment on the JavaLobby thread ), i have decided to take my test Swing application for a ride.
This application is a test-bed for Substance LAF, so it utilizes a lot of Swing-related stuff, creating all possible core components, a lot of listeners, various layouts, colors etc. After downloading the latest release of JRockit 5.0 and installing it, here are the results (first of all i should mention that i didn't get any compilation or runtime errors, so i guess that "write once run everywhere" is still partly relevant).
There are two phases that i had timed - the UIManager.setLookAndFeel function (which eventually calls the constructor of some LAF, but also does a lot of other stuff) and the creation of the main frame itself (constructor, pack, setting the size, location and visibility). The results are:
- For JRockit R26.0.0 version 5.0_04 - 1890 ms for initializing the LAF, 3690 ms for showing the main frame and 22MB taken on the heap.
- For Sun VM version 5.0_06 - 1200 ms for initializing the LAF, 1430 ms for showing the main frame and 7MB taken on the heap.
- For Sun VM version 6.0 b63 - 840 ms for initializing the LAF, 1500 ms for showing the main frame and 8MB taken on the heap.
The results are for single-CPU (2.4GHz) running WinXP SP2.
As usual, the results of such a 'benchmark' should be taken with a shovel of salt, since you may have different LAF and different Swing application altogether. However, the main conclusion is that JRockit may be 'smoking' hot for running the WebLogic, but it's certainly not a prime option for running desktop applications.