Hi Mark, thanks a lot for the clarification, and the hint with the indexes. Regards Martin Mark W. Farnham schrieb: > UNDO_RETENTION is about time, not size. So if the time to complete the impdp > including the index creation is longer in duration that UNDO_RETENTION, then > Oracle's RDBMS has been permitted to release rather than extend UNDO once > that amount of time has gone by, even if the retention column of > DBA_TABLESPACES for the relevant UNDO tablespace is GUARANTEE. > > This, in turn, means that the relevant select may not be "huge," just long > running. I'm not sure WHY impdp has such a long running query on changing > underlying data, but making UNDO_RETENTION longer may well preserve this > information with relatively little addition to total UNDO. Having a small > UNDO_RETENTION just allows Oracle to skip adding more and re-use the expired > (that is older than retention and not involved in an active transaction) > bits of undo. > > That said, it seems likely that running impdp without creating the indexes > and then creating the indexes separately could dramatically decrease the > time of the impdp run and possibly can provide you with faster completion of > the overall job, depending on how the indexes are then added. > > Regards, > > mwf > -- Usn's IT Blog for Linux, Oracle, Asterisk http://www.usn-it.de -- //www.freelists.org/webpage/oracle-l