I would bet you see a performance decrease as oracle computes which hash partition you need to insert/update/delete into. I've experimented a few times with hash partitions (subs included) and always went back to range or list with no subs. I think hash are really suited to selects in order to spread i/o. I'm assuming you've exhausted the other partitioning methods. DML on partitions that don't follow the partitioning strategy are alway a pain in the ass. Mike ganstadba@xxxxxxxxxxx ----- Original Message ----- From: <ryan.gaffuri@xxxxxxx> To: <oracle-l@xxxxxxxxxxxxx> Sent: Friday, April 30, 2004 10:27 AM Subject: anyone ever use sub-partitioning? > We have a table which is about 8GB with about 8 billion records. We add deltas to the file daily(inserts,updates,deletes). > > We are considering paritioning this table so we can use parallel DML. We are considering partitioning on the primary key to about 10-20 partitions. > > Has anyone used hash sub-partitioning? What are your experiences with that? I'd like to have granularity smaller than 400-800 MB per partition. > > ---------------------------------------------------------------- > Please see the official ORACLE-L FAQ: http://www.orafaq.com > ---------------------------------------------------------------- > To unsubscribe send email to: oracle-l-request@xxxxxxxxxxxxx > put 'unsubscribe' in the subject line. > -- > Archives are at //www.freelists.org/archives/oracle-l/ > FAQ is at //www.freelists.org/help/fom-serve/cache/1.html > ----------------------------------------------------------------- > ---------------------------------------------------------------- Please see the official ORACLE-L FAQ: http://www.orafaq.com ---------------------------------------------------------------- To unsubscribe send email to: oracle-l-request@xxxxxxxxxxxxx put 'unsubscribe' in the subject line. -- Archives are at //www.freelists.org/archives/oracle-l/ FAQ is at //www.freelists.org/help/fom-serve/cache/1.html -----------------------------------------------------------------