Hi Christian,
Op 19 aug 2010, om 13:18 heeft Christian Grün het volgende geschreven:
The stack trace indicates that the merge of several temporary index instances causes problems. Regarding the SVN history, the current index builder shouldn't have been changed that much since… 6.1.8? As a shortcut, it might help to give the JVM different -Xmx values; 1gb should be completely more than enough for 780 mb of input data. In the GUI, you should be able to observe how much RAM is actually used during index construction.
That was it. 1Gb didn't work, but 2Gb was enough to get it done in about 50 seconds.
I don't get anything else, if there's something I can do to debug it let me know.
It might help if you could provide us with the original data (which will of course be handled confidentially). It might take some time, however, as I'm currently busy writing my thesis…
I can't supply you with the original data I'm afraid. That would breach my contract with the MPI.
Anyway, thanks for your help again and good luck with the thesis.
Huib.