I had EXACTLY this situation - in my case the production table was holding more than 60 millions of rows, and I needed search in all of them, finding the smaller resultset and putting it in another table (GTT, to reduce log/rollback) - and read consistency was not a issue in my case (the table was VERY static in nature).
Of course, read ALL this enormous table directly in o one-off way WILL stay processing for a while, put a heavy load in I/O sub-system, burn a lot of CPU... So , my goal was : read a number of lines (say, 100000), put them in a temporary table, close cursors, release resources and WAIT for some minutes (giving chance to the machine work for my users), then read the ** next ** 100000 rows, and so on..
I was successfull in this approach using Tom Kyte´s method as showed in http://asktom.oracle.com/pls/ask/f?p...D:127412348064

[]s

Chiappa