[jdom-interest] Thread questions regarding JDOM SAXBuiler?

[David Wall]

That's really interesting, though it also appears you've set some GC
arguments as well (I'm not familiar with them).  But the GC did sound like
it might be a problem, though in the end, they both should have created the
same large XML document objects that would then be discarded.  But this
continues to show 'reuse==true' creates a performance hit.  Why have an
option that appears to slow things down, and even the javadocs indicate it
would slow things down.  What's the purported benefit of doing reuse==true?

David


----- Original Message ----- 
From: "Laurent Bihanic" <laurent.bihanic at atosorigin.com>
To: "Per Norrman" <per.norrman at austers.se>
Cc: "David Wall" <d.wall at computer.org>; <jdom-interest at jdom.org>
Sent: Wednesday, September 01, 2004 7:54 AM
Subject: Re: [jdom-interest] Thread questions regarding JDOM SAXBuiler?


>
> I don't think your test tool performs enough iterations for the result to
be
> reliable due to garbage collector side-effects.
> I modified your tool to run 1000 iterations instead of 20 :
>
> java -XX:+AggressiveHeap large.ThreadedReader
>
> Reuse=true      size=21837      time: 42497
> Reuse=false     size=21837      time: 40796
>
> Reuse=true      size=1927       time: 3403
> Reuse=false     size=1927       time: 11125
>
> Reuse=true      size=21837      time: 28469
> Reuse=false     size=21837      time: 35527
>
> Reuse=true      size=43469      time: 57238
> Reuse=false     size=43469      time: 63963
>
> Reuse=true      size=325632     time: 428557
> Reuse=false     size=325632     time: 443638
>
> ...
>
>
> Per Norrman wrote:
>
> > Hi,
> >
> > I meant to make the program self-cotained but missed the dependency
> > on the concurrent jar. Here's a new version. You should run the test
> > in your environment to confirm the results.
> >
> > Yes, documents are discarded after being built. There are many
variations
> > you can do in a test like this. My guess is that it's
String/StringBuffer
> > handling in SAXBuilder and/or Xerces that accounts for the resuts.
> >
> > A typical output in my environment (P3, 850Mhz, Dell Latitude C600):
> >
> > Reuse=true    size=21731    time: 5720
> > Reuse=false    size=21731    time: 2215
> >
> > Reuse=true    size=1918    time: 200
> > Reuse=false    size=1918    time: 300
> >
> > Reuse=true    size=21731    time: 1200
> > Reuse=false    size=21731    time: 2065
> >
> > Reuse=true    size=43259    time: 3697
> > Reuse=false    size=43259    time: 2663
> >
> > Reuse=true    size=324070    time: 25435
> > Reuse=false    size=324070    time: 22233
> >
> > Reuse=true    size=756109    time: 66417
> > Reuse=false    size=756109    time: 53194
> >
> > The first run should be disregarded. Used for warming-up.
> >
> > /pmn