With my changes to FOP I can process tens of thousands of pages of
XML:FO in only a few Mb of heap. I have asked for testers but so far
noone has responded... this sounds like an ideal environment in which to
test, no? The performance of FOP seems also to have improved, probably
because the GC is a lot less stressful (far fewer objects to scan and
far fewer OS memory requests being performed).
I hope to have my final changes completed today at which time I will
post a summary of them and submit them for inclusion... they are not too
major. I have modified all of the renderers to suit and have a slightly
different API and calling sequence. But I'm happy to send a JAR and the
sources to anyone who wants them earlier.
One of the main problems with FOP IMHO is not that it is poorly designed
but that there has been little control over the quality of the code that
is present there. There are wild variations in the code, assumptions
made in it, commenting, use of public/private members, coupling,
cohesion, maturity, you name it and it changes from file to file and
module to module. I make this observation because a redesign of FOP is
not going to make these issues go away - another solution is required.
I'll look at the synchronization issues once I'm finished with the
pipelining issues, assuming my patches are worthwhile :)
Weiqi Gao wrote:
>We did some playing around with a simple FOP servlet that does
>XHTML->XSL-FO->PDF using Xalan and FOP and come away with the following
>1. It uses quite a bit of memory. While running around 20-30 concurrent
>users, each generating a set of four PDFs ranging from 2 to 5 pages five
>times each, the servlet engine memory usage is about 110MB.
>2. It uses quite a bit of CPU. While running the above test, the CPU is
>about 50% with 10 users, 70% with 20 users, and over 80% with 30 users.
>3. It takes quite a bit of time to do the generation. The minimum
>generation time for the four documents are from 1 to 4 seconds (the 4
>seconds one is 5 pages with 13 tables with various cell colors and
>border styles). In the 30 user test, the average time are from 20 to 40
>seconds, and maximum well over 2 minutes.
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, email: [EMAIL PROTECTED]