users@jaxb.java.net

strange object creation behavior

From: Cliff Martin <cliff_at_lucent.com>
Date: Tue, 30 Nov 2004 15:47:45 -0500

I'm using jaxb 1.0.3 with jwsdp-1.4 and have encountered some strange runtime behavior.

I've got a program that builds a bunch of jaxb objects which the calls the marshaller to
output to a file. I'm using java 1.4.2 sdk to build/run the program.

I started running the application on a new windows xp pro box and noticed that it seemed
to hang at the point of generating the xml document. I did some testing and discovered
that a lot of time is being spent creating the jaxb objects (objectFactory.createXXX
xxxSetYY, etc). I can run the app on a FreeBSD box (also java 1.4.2) and it takes a few
milliseconds to create each object. The processor is a 600 MHz PIII with 768 MB of RAM.
The objects seem to be created quickly with a fairly uniform time between objects,
although each object takes a slightly different time.

I ran on a box with XP Home, 512 MB ram and 1GHz PIII processor and it tends to create a
bunch of objects, stop for a second or so, create more etc. The time for creation/file
generation is satisfactory though.

The one that is causing me trouble is a dual 3GHz P4 processor machine with 2GB of ram and
Windows XP pro. On this machine, each object creation takes 4.5 seconds. The time is
pretty much constant regardless of what object gets created. Strangely, the CPU is pretty
quiescent. I don't see garbage collection going on - it's almost like the app is sleeping
or something. There is nothing going on except calls to the jaxb generated code. The
only real difference is that this machine has to be accessed over a vnc channel, but that
doesn't seem to affect any other applications on that machine. I've tried forcing it to
run on one CPU as well with no change in behavior.

The application also runs fine on some sparc/solaris boxes.

Classpaths, jarfiles, invocation arguments are the same on each platform. I'm at a loss
to explain how the most powerful system by far has such abysmal results.

Has anyone seen anything like this before?? I'd appreciate any insights anyone may have.


Thanks,

Cliff Martin
cliff_at_lucent.com




---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe_at_jaxb.dev.java.net
For additional commands, e-mail: users-help_at_jaxb.dev.java.net