Using HSSF, a simple test of 10000 rows by 255 columns runs fine with "-Xms128m -Xmx394m" and is fast. Using XSSF, the same exact test fails before even getting to 1000 rows, and is extremely slow. After bumping the memory to "-Xms512m -Xmx2048m", it fails somewhere between 3000 and 4000 rows. Please see the attached sample code, which is based on your Timesheet Demo. Thank you, Rob
This may not be a Macintosh-only problem. I am hoping to test on another platform shortly.
Just tested with the same result under Windows Vista, so this does not appear to be a platform-specific issue.
Created attachment 23317 [details] Sample code that illustrates the bug I am resubmitting my original attachment (unchanged), as it isn't appearing in the bug. If there is a delay and I've submitted it twice, I apologize.
XSSF is xml based, so processing the files will always take more memory than using HSSF. Also, in the interests of developer time, we use XML Beans, which allows faster development at the expense of more memory used. If this is proving to be a problem for you, do please do some profiling to identify the heavy memory use areas, and contribute back patches to reduce the memory use!
There was a discussion about it some time ago. See http://markmail.org/thread/vqut6wy3ashguz6x A possible workaround is to stream your data directly in XML. See an example demonstrating my idea: http://svn.apache.org/repos/asf/poi/trunk/src/examples/src/org/apache/poi/xssf/usermodel/examples/BigGridDemo.java Yegor