Performance comparison of Substance skins

June 18th, 2008

As a follow up to the previous post that showed how LightBeam tool can be used to analyze the performance of various core and third-party Swing look-and-feels, today i’m going to answer Luke’s question in the comments section and compare the performance of the core Substance skins.

The table below lists all available core Substance skins, and three numbers for each skin. The first number is the average time to complete a static performance scenario (rendering a bunch of controls to an offscreen image). The second number is the heap size after the static performance scenarios have been completed. The third number is the total time to run all dynamic performance scenarios (the list can be found in the PDF report or in the source code).

There is a simple explanation why Magma, Office 2007 family and Raven family have 25-45% worse dynamic performance than the rest – they use watermarks. I have already mentioned that this has been identified as the biggest performance obstacle during the performance tuning of version 5.0dev, and the skins that used the simple translucent-fill watermark do not use any watermark at all. The selected skins mentioned above are still using non-null watermarks to provide better visuals. You can extend these skins and set the watermark field to null in the constructor to get better performance (but less distinct visuals).

As you can see, there are minimal differences in the rest of the skins, even though they use different button shapers and painters. The core button shapers and painters have been improved to provide uniform performance so that you don’t have to sacrifice the visuals for the performance (except, once again, the watermarks).

If you do find that your custom skin has distinctly different results under different painters, i would be very interested in hearing about this in the project forums, mailing lists or a private e-mail.