Pfff...sorry for the noise.
I'll double check my measurements on MacOS and will also try to
measure on Debian (where the app is in production). I can live with
the slowdown on Mac and so hopefully it does not show up on Debian
Thanks again!
Peter
On 13 Apr 2008, at 11:19, Jeremias Maerki wrote:
That's why I provided two different measurements for the 0.95 branch:
once with JAI's codec and once with Commons' TIFF codec ("internal
TIFF
codec") which is enabled by my fix from Friday. You can see that the
internal TIFF codec adds about 0.4 seconds in my environment
compared to
the JAI codec.
On 13.04.2008 10:38:27 Peter Coppens wrote:
Thanks Jeremias....as always, very thorough work/analysis.
Is it correct you always used JAI TIFF codec in your measurements and
never the xmlgraphics fix you added on Friday. The one that supports
tiff without jai at all?
If so, any chance at all you could measure the performance of that
fix/
patch on your xp box as well?
Thanks,
Peter
On 13 Apr 2008, at 10:02, Jeremias Maerki wrote:
Peter sent me one of his large TIFF images to test with. Thanks,
Peter.
I used his FO snippet he posted at the beginning of the thread so
the
firgures below are including Batik into the picture.
The TIFF file: RGB, LZW compression, 1327x3682 pixels, 2.73MB
I couldn't see a factor 6 increase of time necessary to process the
picture on my Windows XP SP2 except when run inside my profiler but
that's due to method instrumentation overhead. The normal runtime
effect
was far smaller. As you can see I didn't experience slower
performance
with FOP Trunk/0.95beta compared to 0.94. Here's what I measured
(please
note that the figures below always refer to the server VM with
parallel
GC enabled, JVM startup and class loading is included in the times):
FOP 0.94 (Commons' TIFF codec)
Java 1.4.2: 4.0 sec
Java 1.5.0: 3.6 sec
Java 6: 3.3 sec
Trunk (comparable to 0.95beta, JAI TIFF codec)
Java 1.4.2: 2.9 sec
Java 1.5.0: 2.9 sec
Java 1.6: 2.6 sec
When profiled with my YourKit profiler I found two hotspots which
are
now fixed:
http://svn.apache.org/viewvc?rev=647537&view=rev
http://svn.apache.org/viewvc?rev=647536&view=rev
The latter hotspot in Commons is only triggered when my new TIFF
image
loader is used.
0.95 Branch (including my performance fixes, JAI TIFF codec):
Java 1.4.2: 2.2 sec
Java 1.5.0: 2.0 sec
Java 6.0: 1.9 sec
0.95 Branch (including my performance fixes, internal TIFF codec):
Java 1.4.2: 2.7 sec
Java 1.5.0: 2.4 sec
Java 6.0: 2.3 sec
Just as an experiment I've converted the TIFF image to JPEG with
relatively low compression so there are no serious artifacts. The
resulting JPEG image is 425KB. If performance is important and the
images are suitable (fotos), JPEG can be an option since the images
don't have to be decoded for PDF production. Here's the result:
0.95 Branch (including my performance fixes):
Java 6.0: 1.1 sec
On 12.04.2008 11:37:01 Jeremias Maerki wrote:
<snip/>
If Peter sends me his test TIFF I'll profile the problem. A
factor 6
increase is certainly not to be expected.
<snip/>
Jeremias Maerki
Jeremias Maerki
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
