Re: FOP Weakness

[Chuck Paussa]

Xie, David (IPCG-NJ) wrote:
Hi All,
I have three questions regarding FOP.
First of all, does anyone know what FOP stands for?  My guess File object 
program.
Second question is regarding the FOP performance.  I notice that every time FOP 
runs on Jakarta Tomcat 4.0, the cpu usage is almost 100%.  This is not good if 
we are running other applications on the
same server.  Anyone experiencing such problem?  Do you have a solution?  Does 
increasing heap size make any difference?
My final question is regarding FOP scability.  I have been doing numerous tests on FOP using Loadrunner, a simulation application that allows us to specify certain number of users and number of
requests.  When making large number of requests to jakarta Tomcat and FOP to generate pdfs, I kept on getting socket write error (java.net.socketException Error: connection reset by peer: socket write
error).  Anyone know what cause this error?  On the web, it said this error is caused by browser and it's not big deal.  

Thanks for you help, I look forward to your expert opinion.
Dave   
 

FOP = Formatting Objects Processor  ;   FO is an 
extension to the XSL standard
Performance, FOP takes a lot of memory. To get 200ppm processing, I'm 
estimating 2GB memory on the system + memory for the database and web 
server. Improvements in the memory footprint are scheduled to be started 
not soon Hey! What do you want? memory is cheap.
Scalability. If you provide it with scalable amounts of memory, the 
performance is fairly linear.

Re: FOP Weakness

[James Richardson]

Chuck Paussa wrote:

Performance, FOP takes a lot of memory. To get 200ppm processing, I'm
estimating 2GB memory on the system + memory for the database and web
server. Improvements in the memory footprint are scheduled to be started
not soon Hey! What do you want? memory is cheap.

Hmm... Interesting estimates. However, while memory is indeed cheap (at 
least on intel achitectures, for enterprise-grade machines it most 
certainly is not). If you are suggesting a 2GB heap size, the garbage 
collection times will kill you.

I'm experimenting with running Jini services to do the processing (or 
maybe a JavaSpace, later), then we can have as many processors as 
required - not one gigantic process

I would also suggest in the future trying to make inherently parallel 
tasks run in multiple threads. I have about 8 processors available to 
me, but FOP appears to run in 1 uber-thread.

Cheers
James