Dimensional Modelling compared to Relational (3NF +) Modelling for DSS

Hi all,

I'm researching how the Kimball and Inmon approaches have developed since inception.

What are peoples views and thoughts on this?

Has either approached softened towards the other?

Have advances in software (for instance in memory analytics) or hardware (large memory systems, SSD's) steered the methodolgies?

Many thanks,
Tony.

I don't think things have changed much. But, you misrepresent Inmon in your heading. The 3NF data warehouse is not for DSS (or BI or even general reporting), it is a repository from which data is published for use in DSS/BI/Reporting systems. Data is published in the form of star schema, cubes, extract files, etc....

In memory analytics has been around since the first spreadsheet... its just that PCs have more of it. It is not clear if in-memory OLAP is particulary better than a MDDB based cube. Both have pros and cons, so it depends on what you prefer. And, as far as solid state disk goes, it is not that big a game changer... performance is not significantly better.

If you roll out a 3NF model for BI, no matter what the platform, you still have a confusing, and cumbersome schema to deal with.... and a lot of work hiding it from end-users.

Inmon definitely recommends using a star schema for data marts. Just not for data warehouses. I can see his point - if you are building a very large data warehouse and you don't have specific requirements other than "Make sure you capture everything" then it makes sense to use a 3NF data warehouse, then load into a Star Schema data mart.

As far as In-Memory, my friend and former colleague Boyan Penev has written an article here: http://www.bp-msbi.com/2011/01/powerpivot-data-modelling-for-performance/

Boyan shows that even with an in-memory database such as Excel PowerPivot, the Star Schema was more performant than a 3NF.

I disagree with the comments below. In his writings and in particular his paper "Corporate Information Factory - The DSS Environment - Data Warehouse, Data Marts and Data Mining, A glimpse at the past, A Peek at the Future"
http://www.inmoncif.com/view/23
Inman makes it clear that he believes the Data Warehouse is the essential (but not sufficient) component of a DSS. It is hard to argue against such a position. The question however is, does the Data Warehouse have to be in 3NF as in the Inman architecture?


.

ngalemmo wrote:I don't think things have changed much. But, you misrepresent Inmon in your heading. The 3NF data warehouse is not for DSS (or BI or even general reporting), it is a repository from which data is published for use in DSS/BI/Reporting systems. Data is published in the form of star schema, cubes, extract files, etc....

In memory analytics has been around since the first spreadsheet... its just that PCs have more of it. It is not clear if in-memory OLAP is particulary better than a MDDB based cube. Both have pros and cons, so it depends on what you prefer. And, as far as solid state disk goes, it is not that big a game changer... performance is not significantly better.

If you roll out a 3NF model for BI, no matter what the platform, you still have a confusing, and cumbersome schema to deal with.... and a lot of work hiding it from end-users.

The key word is "component" of a DSS. The data warehouse is not intended to be used directly by the DSS system, but serve as a data source for loading an appropriate data structure to support the DSS application.