Data Mart Does Not Equal Data Warehouse

Alan Musnikow Wed Dec 28, 2011 11:03 pm

For an unexplained reason, the 27-Dec-2011 issue of Information Management Daily included a link to Bill Inmon's May 1998 article, "Data Mart Does Not Equal Data Warehouse."

I posted the following comment.

You have built a straw man data mart concept and then shown that it is flammable.

Conformed dimensions in data marts enforce the need for "need for ... organizational ... discipline and ... concern for the long-term architecture that is created by the data marts."

There is no reason why data "found in data marts" should be "highly indexed." In fact, data marts built on a columnar database such as Infobright have no indexes.

There is no reason why data marts need to be organized by department rather than "arranged around the corporate subject areas found in the corporate data model."

There is no reason why data marts cannot be "built and owned by centrally coordinated organizations, such as the classic IT organization."

There is no reason why data marts cannot contain "the most granular data the corporation has."

There is no reason why data marts cannot contain "a robust amount of historical data."

There is no reason why "data mart granularity" should be any different "than that found in the data warehouse...."

Jeff Smith Thu Dec 29, 2011 3:53 pm

Immon has a different approach than Kimball and he's a little touchy about it to. No one is going to convince Immon.

I don't agree with the approach. It might have made sense with performance was a challenge, but technology has improved so much. The whole 1 database with multiple schemas VS multiple database on 1 instance VS multiple databases on multiple instance on 1 server, vs multiple servers has to be thought out. I'm of the mind that 1 database with different instances is the way to go.

But I can see the benefit of a stand alone data mart for people who need to be able to write SQL directly against the data without interferring with users running reports. Nothing like a cartesian product to increase the volume of IT tickets complaing about report performance.

But even in that case, I'd prefer to see the datawarehouse cloned so that everyone is still working from the exact same dataset. Cut's down on development as well.

ngalemmo Thu Dec 29, 2011 4:51 pm

There are also a lot of different definitions as to what a data mart is. In Inmon's world a data mart is a very specific focused subject area targeted for specific analysis and is usually an aggregate. Marts like that cannot be integrated to form a proper data warehouse.

In fact, data marts built on a columnar database such as Infobright have no indexes.

Not to split hairs, but columnar databases are essentially collections of bitmap indexes. While you may not declare indexes, the underlying data structures are nothing but indexes. The only database system that I know of that does not have indexes is Netezza, which is a row oriented MPP appliance (although they do have zone maps which is a block based filtering method that helps performance).

Sponsored content

Data Mart Does Not Equal Data Warehouse

Data Mart Does Not Equal Data Warehouse

Re: Data Mart Does Not Equal Data Warehouse

Re: Data Mart Does Not Equal Data Warehouse

Re: Data Mart Does Not Equal Data Warehouse