Re: partitioned table advice

OK, this is the kind of feedback I was looking for. Thanks.

On Fri, 14 Dec 2001 02:26:41 GMT, Mark Townsend <markbtownsend_at_attbi.com> wrote:

>in article 3c193f41.87010724_at_ausnews.austin.ibm.com, Ed Stevens at
>Ed_Stevens_at_nospam.noway.nohow wrote on 12/13/01 3:59 PM:
>
>> Platform; Oracle 8.0.5 SE, NT4
>>
>> We've never used a partitioned table before, but I'm looking at a candidate.
>> A
>> SELECT DISTINCT on the primary key shows seven different values. The
>> distribution of rows across these values is as follows:
>>
>> value 1 - 17,814 rows
>> value 2 - 322 rows
>> value 3 - 36,976 rows
>> value 4 - 29,580 rows
>> value 5 - 228 rows
>> value 6 - 1,565 rows
>> value 7 - 371 rows
>>
>> The app is heavy in queries with a "WHERE =" on this column.
>>
>> Would the skewing of the column values argue for or against partitioning on
>> this
>> column? Or is the answer still "it depends -- we need more info about which
>> distinct values are most often queried."
>>
>>
>> --
>> Ed Stevens
>> (Opinions expressed do not necessarily represent those of my employer.)
>
>
>1) You're on SE - partitioning is an Option to EE, so it will cost twice
>(once to upgrade to EE, the again for the partitioning option)
>
>2) Partitioning for data management is overkill - you simply don't have
>enough data to make partitioning worthwhile
>
>3) It's not clear what your actual problem is, but if it's performance
>differences between queries on the small value sets versus the large value
>sets, then look at indexing and even potentially histograms (presuming you
>are using the CBO for the latter, of course). See
>http://otn.oracle.com/docs/products/oracle8i/doc_library/817_doc/server.817/
>a76992/stats.htm#27027 - but also check that histograms are an SE feature.
>

--
Ed Stevens
(Opinions expressed do not necessarily represent those of my employer.)