You are viewing a plain text version of this content. The canonical link for it is here.
Posted to fop-dev@xmlgraphics.apache.org by Lukas Pietsch <Lu...@pluto.uni-freiburg.de> on 2001/12/11 17:59:30 UTC

FOP performance on Win98/JRE 1.3.1

Hello,

(I asked this question on the www-xsl-fo@w3.org list first, but was told to better go here with it.)

I'm new to xsl-fo and related matters and I've just managed to set up FOP (0.20.2) and make a few test runs. I'm now wondering a bit about its performance. Parsing a single test xml document (a docbook chapter) with just a few words in it, transforming it with the docbook xsl stylesheets and outputting to pdf takes more than 40 seconds. It's fifteen seconds before I even get the debug line ("[DEBUG]: using SAX parser org.apache.xerces.parsers.SAXParser"). Afterwards, it says "[DEBUG]: Avg render time: 10600ms/page".
Is this considered normal? It makes me wonder if my FOP installation (or indeed my Java installation) is set up correctly. Cyril Rognon on the www-xsl-fo list confirmed he found it rather too slow.

This is on Windows 98, with a 500MHz AMD-K6 processor and 128 MB memory, using FOP 0.20.2 (sorry, I wrongly said 0.20.0 in the earlier mail) and the Java JRE 1.3.1.

What could I do to diagnose the problem better?

Thanks for any pointers,

Lukas Pietsch






---------------------------------------------------------------------
To unsubscribe, e-mail: fop-dev-unsubscribe@xml.apache.org
For additional commands, email: fop-dev-help@xml.apache.org


Re: FOP performance on Win98/JRE 1.3.1

Posted by Lukas Pietsch <Lu...@pluto.uni-freiburg.de>.
Thanks, James, for the detailed comments. Not that I understood everything of it--I'm afraid I'm rather unexperienced with Java in general. Your suggestions sound quite convincing, only I don't know how to actually carry them out. Now maybe what follows is terribly boring newbie stuff. In that case, perhaps somebody could point me to some relevant tutorial or similar stuff on the web?

(1) making the JVM survive a single FOP run or a single document conversion. How? The only way I know of invoking FOP is by saying "java org.apache.fop.apps.Fop" in a .bat program.

(2) Adjusting JVM memory settings. How? Okay, I've found something about -Xms and -Xmx commandline parameters, but how do I find out what the present default values are?

(3) Eden heap space settings. No idea what those are, let alone how I could change them...

(4) Increasing physical memory. Well, okay, I do understand how I could achieve this--just spend a couple of hundred bucks. ;-) But are you really suggesting it's hopeless to run FOP on a 128MB machine?

Lukas



---------------------------------------------------------------------
To unsubscribe, e-mail: fop-dev-unsubscribe@xml.apache.org
For additional commands, email: fop-dev-help@xml.apache.org


Re: FOP performance on Win98/JRE 1.3.1

Posted by James Richardson <ja...@db.com>.
Just a thought... did anybody try FOP with jRockit JVM?

James


---------------------------------------------------------------------
To unsubscribe, e-mail: fop-dev-unsubscribe@xml.apache.org
For additional commands, email: fop-dev-help@xml.apache.org


Re: FOP performance on Win98/JRE 1.3.1

Posted by IvanLatysh <iv...@yourmail.com>.
Hello, James!
You wrote to <fo...@xml.apache.org> on Tue, 11 Dec 2001 17:28:25 +0000:


 JR> The performance of FOP needs quite a bit of work. At least thats the
 JR> impression I get from many people round here. However, its really
 JR> new  software ( hanve version 0.20 I would imagine ), and I think
 JR> that that  probably the coders are working on getting the
 JR> functionality there  before spending time concentrating on the
 JR> performance.

I think performance only question of time.
But one more:
What about German documentation for some classes.
Maybe you guys could notify all developers to use English.
Because "I am stack in a middle with you" (c) famous song.
I couldn't move from dead point with my application.
I do understand that now some thing are not ready yet, 
but I think you should pay attention to preview panel and print method.

And, by the way - great job guys.

---
Yours sincerely, Ivan Latysh.
Ivan@yourmail.com
http://ivan.yourmail.com


---------------------------------------------------------------------
To unsubscribe, e-mail: fop-dev-unsubscribe@xml.apache.org
For additional commands, email: fop-dev-help@xml.apache.org


Re: FOP performance on Win98/JRE 1.3.1

Posted by James Richardson <ja...@db.com>.
The performance of FOP needs quite a bit of work. At least thats the 
impression I get from many people round here. However, its really new 
software ( hanve version 0.20 I would imagine ), and I think that that 
probably the coders are working on getting the functionality there 
before spending time concentrating on the performance.

One thing you will find is that the first run through the code takes by 
far the longest time. The JVM needs to load all the relevant classes, 
initialise them, and that kind of thing. Once this has been done, the 
JVM (if you have it switched on) will potentially compile some of the 
code, rather than interpret the java byte code. This can improve the 
speed significantly.

The other thing you need to check is for your JVM memory settings. If 
you use the default, then FOP may cause the free heap space to go 
dangerously low, increasing the frequency of full GC runs. These are 
best avoided, as all threads are suspended during the GC phase. Adding 
memory to the heap can cause these problems to diminish, (perhaps adding 
up to 256m, using the -Xms and -Xmx flags ), but then you may find that 
full GCs take too long ...You may also want to take a look at the Eden 
heap space settings, these could significantly increase speed if 
slowdown is due to memory & GC issues, as a result of short-lived objects.

Of course if you are adding memory to the JVM heap, then you should also 
make sure that you have anough physical memory, or this may cause VM 
paging at the OS level. I would say 256 or 512M as minimum requirement.

When running on UNIX i have seen a significant amount of kernel CPU 
usage, which i cannot really account for....

In short, then. If you want FOP to run more quickly, render more than 
one document during the JVM lifetime. The first rendering will be slow, 
but subsequent ones much more speedy. And add memory.


Hope this helps

James




Lukas Pietsch wrote:

> 
> 
> 
> Hello,
> 
> (I asked this question on the www-xsl-fo@w3.org list first, but was told to
> better go here with it.)
> 
> I'm new to xsl-fo and related matters and I've just managed to set up FOP
> (0.20.2) and make a few test runs. I'm now wondering a bit about its
> performance. Parsing a single test xml document (a docbook chapter) with
> just a few words in it, transforming it with the docbook xsl stylesheets
> and outputting to pdf takes more than 40 seconds. It's fifteen seconds
> before I even get the debug line ("[DEBUG]: using SAX parser
> org.apache.xerces.parsers.SAXParser"). Afterwards, it says "[DEBUG]: Avg
> render time: 10600ms/page".
> Is this considered normal? It makes me wonder if my FOP installation (or
> indeed my Java installation) is set up correctly. Cyril Rognon on the
> www-xsl-fo list confirmed he found it rather too slow.
> 
> This is on Windows 98, with a 500MHz AMD-K6 processor and 128 MB memory,
> using FOP 0.20.2 (sorry, I wrongly said 0.20.0 in the earlier mail) and the
> Java JRE 1.3.1.
> 
> What could I do to diagnose the problem better?
> 
> Thanks for any pointers,
> 
> Lukas Pietsch
> 
> 
> 
> 
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: fop-dev-unsubscribe@xml.apache.org
> For additional commands, email: fop-dev-help@xml.apache.org
> 
> 



---------------------------------------------------------------------
To unsubscribe, e-mail: fop-dev-unsubscribe@xml.apache.org
For additional commands, email: fop-dev-help@xml.apache.org