Heiko, The reason you are having performance problems is that Inline::Java does not actually transfer the array over to Perl. It actually sends a 'reference' to it an calls Java each time you want to access an element. In your case you have 4000 * 50 = 200000 elements, so that's why it's slow.
Here are things your could try: - More Perl-Excel stuff: http://search.cpan.org/search?m=all&q=Excel&s=1 - Create a Java wrapper class that will return each row (as a tab separated String), instead of each cell. That way you can reduce your number of Perl-Java trips by 50. - Use a temporary ASCII file that Perl can parse easily afterwards. > Hi, > > I have to read large Excel-files (5sheet, 4000rows, 50columns) with > perl. Misereably, the Spreadsheet::ParseExcel modules is really slow and > memory hungry on this (> 5min parsing, > 250MB memory). > > Therefore I've been looking around and found the nice jakarta POI > project, handling exactly the same problem, but in Java. Parsing is > quite fast (~30s, 100MB) on the Java process. It was next to no problem > to attach this to perl using Inline::Java. > > My problem is getting the data back from java to perl. I tried to > transfer the data at once, so I created a String[][] per sheet and get > it by a function to perl. But when I try to use the values, it is still > accessing the Java process, and while parsing takes only 30s, accessing > each value of my String[][] of one sheet takes ~2min (about 70% perl and > 30% java cpu usage (top)). Using JNI reduced the time but it's still > above 1min. > > > Has anybody a good idea how to transfer the data with only one call to > the JVM? Any ideas about other projects reading Excel-files by Perl are > welcome, too. (Dropping excel is not possible, misereably.) > > Best regards, > > Heiko > >
