I have been tracking down performance issues in optimizing MiniGUI peers for AWT.

We are doing pretty well with drawing speed, but we found that Container was very very slow handling the paint method
This we tracked down to the MiniGUI peer code, specifically, that it was creating a lot of new instances of various objects in Graphics.create.

The performance impact of the object creation was very very high - much higher than you would expect in terms of processor speed proportions.

That, and some searches for benchmark results on Kaffe GC lead me to suspect that memory allocation is a performance bottleneck in Kaffe.

I was wondering if anyone had similar experiences?

So, while we optimize paint performance using the fact that Graphics.create and Graphics.dispose can be used to manage a cache of Graphics instances, and to initialize those instances instead of using "new" (often gratuitously, as it turned out, for things like Color), we think the performance of Kaffe GC might be a general drag on performance.

We tried using Boehm GC, but encountered an issue which is now bug #113. Any experiences/pointers getting Boehm GC up on ARM Linux would also be very useful.

ZM
--
Zigurd Mednieks
Chief UI Architect
Mobile: +1 617 447 3595
Email: [EMAIL PROTECTED]
IM: zigurdm (Skype)
http://www.linkedin.com/in/zigurd
The 4th Screen
D2 Technologies
2 Clock Tower place
Suite 360
Maynard, MA 01754 USA
The leader in embedded VoIP software
See who we know in common Want a signature like this?
_______________________________________________
kaffe mailing list
[email protected]
http://kaffe.org/cgi-bin/mailman/listinfo/kaffe

Reply via email to