On Apr 11, 2008, at 23:30, Peter Coppens wrote: Hi Peter
I don't like the jai dependancy (don't like native Java libraries)
Very rightly so. As demonstrated by this particular case, if one platform lacks native support, it seriously breaks the usability of the library as a platform-independent way of handling images.
Maybe this will be better in Apple Java 6, but I can't test since the DP is only available for 64-bit Intel Macs and I'm still on PowerPC... :(
but Jeremias' patch would solve that. The 6x performance hit is ....more troublesome :(
Does this performance hit here also increase with the number of TIFFs you include (as you verified with Sun's pure Java implementation), or is the effect restricted to the first one? If you include, say 6 different images, is there still a 6x performance hit, or does it shrink? The latter would indicate that the performance hit could actually be a sign of good scalability (always think positive ;-)). A performance /hit/ for very low numbers, which decreases, then once a threshold is reached, translates itself into an actual performance / benefit/ for the larger numbers (?) An algorithm with linear behavior will perform better (sometimes / much/ better) for only one execution, than one with logarithmic behavior. The latter, OTOH, will very soon become your friend as the number of executions starts to grow. Slower for a single execution due to the added complexity, but this added complexity would precisely have as its purpose to reduce the runtime for future executions. Only if there are none, does this become pure overhead.
Another thing you could try out: Is there a similar difference in timing if the TIFF is included as an fo:external-graphic? That would already allow to factor Batik out of the equation.
Cheers Andreas --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
