Re: multi-column indexes, statistics and selectivity
Date: Sun, 6 Dec 2009 20:58:51 +0000 (UTC)
On Sun, 06 Dec 2009 13:33:06 +0100, Robert Klemme wrote:
> On 05.12.2009 22:35, Mladen Gogala wrote:
>> On Sat, 05 Dec 2009 14:18:38 +0100, Robert Klemme wrote:
>>> Are there cases where it makes sense to use a FBI to get "multi >>> column" histogram information?
>> Hmmm, let me quote from the 11g book: if the columns are related in the
>> way that breaks the 3rd normal form, such statistics would make sense.
>> The famous example by Jonatan was about zodiac signs and months of
>> birth. The 10g CBO would produce an incorrect estimate for the number
>> of people with the sign of Capricorn born in June. There are, of
>> course, none, but you can't know that without a multi-column histogram.
> > That makes a lot of sense! Thank you, Mladen! > > Kind regards > > robert
Give the credit where the credit is due. The original source is Jonathan Lewis, not me. I do confess to reading his articles vewy, vewy cawefully. I was unable to find the example quickly so here is where Tom Kyte also invokes the same famous example: http://tinyurl.com/yjw5gbeReceived on Sun Dec 06 2009 - 14:58:51 CST