RE: Partition Statistics
- From: "Christian Antognini" <Christian.Antognini@xxxxxxxxxxxx>
- To: <jaromir@xxxxxxxxxxxx>
- Date: Thu, 9 Dec 2004 17:39:27 +0100
Hi Jaromir
>In case when all indices on the partitioned table are defined local =
there is
>really not very much "value added" in the global statistics. (Of course =
in
>case when e.g. the local density of some columns is low but the global =
is
>high this is not true).
In my opinion statistics at table level should be gathered. In fact I =
see no problem doing it and this can only be beneficial for the =
optimizer (e.g. when no partition pruning is used).
>Mathematically seen there is no possibility to calculate all global
>statistics from local (partition) statistics - particularly the =
density. But
>if the above condition are in place, it isn't a big task. I assume =
there is
>an algorithm for default global statistics for partitioned table in =
case
>that local statistics are defined. I assume this could be something =
like
>"get statistics from one (or more) partition(s) and do a little math =
on it"
As you wrote it's not possible to compute table statistics starting from =
partition statistics. Therefore Oracle has introduced global statistics =
in DBMS_STATS (please don't mix the concept of global statistics with =
the statistics at table level... here I'm speaking from global =
statistics).
>I observed relatively precise CBO estimation on queries of a =
partitioned
>table with local statistics only.
For queries with or without partition pruning?
Chris
--
http://www.freelists.org/webpage/oracle-l
- Follow-Ups:
- Re: Partition Statistics
- From: jaromir nemec
Other related posts:
- » Partition Statistics
- » RE: Partition Statistics
- » Re: Partition Statistics
- » RE: Partition Statistics
- » RE: Partition Statistics
- » Re: Partition Statistics
- Re: Partition Statistics
- From: jaromir nemec