Re: Partition strangeness

Hi,

> We've got a BUNCH of partitioned tables, but normally we add one month
> at a time, load it and analyze. Never had any problems.

If you use ANALYZE TABLE .. PARTITION (..) the table statistics are not populated. You may quick review it looking at all_tables view for the partitioned table.

> When we tried 200412, the last of the old partitions, it was over 12
hours.

I expect you use insert select for populating the new partition and you reference the partitioned table in the select. If so, do you reference all partitions or only some of them? In the former case (i.e. there is nor restriction on the partitioning key in the select) the (missing) table statistics are used and should be gathered. You may experiment if some lower estimating factor is sufficient.

Another possibility are the *derived* statistics (global_stats='N') using dbms_stats. But I have no experience with this feature in Oracle 8. And above all you must verify if the aggregation algorithm fits your data.

HTH Jaromir D.B. Nemec

http://www.db-nemec.com

• Original Message ----- From: "Kline.Michael" <Michael.Kline_at_SunTrust.com> To: <oracle-l_at_freelists.org> Sent: Friday, January 28, 2005 12:05 AM Subject: Partition strangeness

--
http://www.freelists.org/webpage/oracle-l