| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
 |
 | |||
| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
|
| |||
Home -> Community -> Mailing Lists -> Oracle-L -> Re: Index help
A key factor is how frequently different values are queried. As Dan says,
if :1= 46847 then a FTS is the way to go. How often is this the case? Or
is the query generally used to find "non-common" values (the other 4%). If
the latter, then I'd recommend using a literal for the IDA3A5 portion of the
query and have a histogram on the column. Same for codeC5; it has 3
distinct values, but if 99% are one value and the query is looking for the
other two, then a literal and histograms are better used here. You need to
look at the query both in terms of the data distributions and the way the
query is being used by the app.
--Terry
If the value of :1 = 46847 a full table scan makes sense (96% of the rows meet the predicate condition). If the value is not 46847, then an index lookup is probably faster. This is really a catch-22. Bind variables are helping your parse times, but *may* be hurting your execution time (hard to tell without knowing how many executions use 46847 and how many do not). Without a histogram on the column, Oracle will assume an even distribution, which is incorrect.
You are trapped between a rock and a hard place. If you use hard-coded values instead of binds and have a histogram on the column, the CBO should pick the proper execution plan each time. This also means that you will have unique versions of the statement in your shared pool. Sorry there is not a simple, quick answer to this one.
I suggest reading Wolfgang's paper "Fallacies of the Cost-Based Optimizer" at www.centrexcc.com. He explains this issue very well.
Regards,
Daniel Fink
M.Godlewski wrote:
> Data distribution.
>
> IDA3A5 count(IDA3A5)
> 4104 44
> 4107 156
> 4110 1
> 4111 1
> 4137 1
> 46847 4905
-- Archives are at http://www.freelists.org/archives/oracle-l/ FAQ is at http://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_at_freelists.org put 'unsubscribe' in the subject line. -- Archives are at http://www.freelists.org/archives/oracle-l/ FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html -----------------------------------------------------------------Received on Mon Jul 26 2004 - 10:46:52 CDT
 |
 |