We saw the same thing. In our case, the culprit is the JVM, GC, and memory 
management. At some threshold (which varies by JVM vendor, JVM settings, etc.) 
it starts thrashing, and performance degrades almost immediately to 
uselessness. We have not found a workaround other than to write a key part of 
our system (the one that absolutely MUST scale) in plain old low-level C, using 
primitive non-validating parsing. 

We used JProbe and dumps from the JVM ( -Xloggc, -Xprof) to help pin it down.

YMMV.

Thomas

-----Original Message-----
From: Conley, Daryl [mailto:[EMAIL PROTECTED] 
Sent: Monday, September 13, 2004 9:55 AM
To: Xerces-J-User (E-mail)
Subject: Processing speed slow down for large files

Hello, 

   We have an application that processes from very small to very large XML 
files.  Up to this point we have not had two many issues with the SAX parser in 
Xerces. Unfortunately we are going to need to process very large files, so I 
have been running some tests to see how the performance is.  I have found that 
I can parse a file up to about 1.15 Gig with a quarter of the records with 
invalid data (to simulate the bad data we will be getting) and it processes in 
about 1.5 hours, a file with the same error rate but 1.2 Gig in size takes 27.5 
hours.  These are running under WebLogic on a Sun dual processor server.  I 
have tested the same file but our process did not output anything because I was 
testing to see if the IO was slowing the
process but it still took 27.5 hours.   What could be the cause of such an
abrupt change in processing time??  Is there any tools that I could use to see 
what is going on under the hood?  By the way a clean file would take about an 
hour to process.

Any help would be appreciated.

Thanks

Daryl

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]


---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to