SQL*Loader on steroids

There is a tight, narrow bottleneck in my site: SQL*Loader. Since all the .dat files are programmatically generated (by my own programs) I have been looking into ways to make those files as lean and predigested as possible. One example: some input files have a large percentage of unique-key duplicates which are later rejected by sqlldr, so I may take steps to never produce a single record in which the uniqueness is violated. I am checking records with bad dates, etc., so they are pre-rejected and long fields are being truncated before going into the .dat file.

All that covers one aspect of the problem.

Where I need help is in another aspect: how else can I make SQL*Loader run faster? I seem to recall some (parallel?, direct?) mode of loading. Is there a substantial performance difference between fixed record length and delimited (the latter is what I currently use)?

TIA, -Ramon F Herrera Received on Sat Mar 31 2007 - 23:37:42 CDT