Re: bitmap index and star transformation must be single colum?

  • From: Greg Rahn <greg@xxxxxxxxxxxxxxxxxx>
  • To: amonte <ax.mount@xxxxxxxxx>
  • Date: Fri, 8 Oct 2010 10:03:43 -0700

Regardless if star transformation works for multi-column keys and column
bitmap indexes, I very much disagree with the comment below to get rid of
surrogate keys.  That makes zero sense.
I won't go as far as to say using multi-column keys and column bitmap
indexes is wrong, but I certainly would not design it that way.  Personally
I use only single column SKs and all of the data molders that I personally
know stick to that as well.  Basic searching on this topic in the Internet
yields the same findings and recommendations.


On Thu, Oct 7, 2010 at 3:26 PM, amonte <ax.mount@xxxxxxxxx> wrote:

> By the way
>
> found in this url where we can see compount bitmap indexes and star
> transformation in action
>
> http://www.business-intelligence-quotient.com/?p=21
>
> Which states:
>
> *This also means that surrogate keys are not a pre-requisite for star
> transformation to be used in a dimensional model. So another reason to get
> rid of them (in most situations).*
>
>
>
>
> Thank you
>
> Alex
>
>
> 2010/10/7 Greg Rahn <greg@xxxxxxxxxxxxxxxxxx>
>
>> I dont think your schema design will work for star transformation.  I
>> believe the requirement is:
>> - primary key on column of dimension table
>> - bitmap index on SK column on fact table
>>
>> In both cases its single column keys.  Given you cant have 2 single column
>> primary keys on your dimension table, it would seem it would not work.
>>
>> In good design there is a single column PK and SK on the dimension and
>> fact table.
>>
>> On Thu, Oct 7, 2010 at 12:25 AM, amonte <ax.mount@xxxxxxxxx> wrote:
>>
>>> Hi
>>>
>>> I am reading 11.1 and 11.2 Data Warehousing Guide and where it explains
>>> about using star transformation with bitmap index in both release it states:
>>>
>>> *A prerequisite of the star transformation is that there be a
>>> single-column bitmap index on every join column of the fact table. These
>>> join columns include all foreign key columns.*
>>>
>>> Is it saying that bitmap index cannot be composite?
>>>
>>> So for example if fact table has FK against dimension such as
>>>
>>> fact.c1 ----> dim1.c1
>>> fact.c2 ----> dim1.c2
>>>
>>> We cannot create a bitmap index on fact.c1 and fact.c2? But we can create
>>> two seperate bitmap index each for each fact table column?
>>>
>>>
>>> Alex
>>>
>>>
>>
>>
>> --
>> Regards,
>> Greg Rahn
>> http://structureddata.org
>>
>
>


-- 
Regards,
Greg Rahn
http://structureddata.org

Other related posts: