Unified Information System

Hi all,
Could you tell me if the following statement is a complete utopia :
we would like to experiment an unified business system where a small business application
is linked directly to a star model. The business logic, read/write/modify data in a fact table
and add dimension's rows if necessary.

Have you ever build such stack ?
What are the pros and cons ?

Best Regards.

Terrible idea. The dimensional model is designed for historical reporting. The application data model should be normalized to provide transaction and data integrity.

How about a 3nf database that feeds the star model on a regular basis.

Yes, but always making etl in order to feed the datawarehouse is very expensive.
In memory db and new technology could make transactional and decisional a more fuzzy frontier.

A simple application using MDM and input form values should insert a fact/process table without data integrity issue.

Am i wrong ?

Thanks !

Yes, you're wrong. Codd developed the concepts of normalization so he could prove, mathematically, that a normalized relational database can reliably and consistently handle transactional processing with multiple concurrent users. A dimensional structure cannot. It has nothing to do with performance or hardware platform, but rather row locking, concurrency and data consistancy.

If you are developing a database to support an application system that allows for multiple, independent, concurrent update processes, the only way to go is with a model in, at minimum, 3NF.

The old school train of though is to then build reporting and query functionality off the 3NF model. However, experience has shown that that approach leads to relatively long running queries that impact OLTP performance. This lead to the data warehousing concept and (later) dimensional modeling.

ngalemmo wrote:Yes, you're wrong. Codd developed the concepts of normalization so he could prove, mathematically, that a normalized relational database can reliably and consistently handle transactional processing with multiple concurrent users. A dimensional structure cannot. It has nothing to do with performance or hardware platform, but rather row locking, concurrency and data consistancy.

If you are developing a database to support an application system that allows for multiple, independent, concurrent update processes, the only way to go is with a model in, at minimum, 3NF.

The old school train of though is to then build reporting and query functionality off the 3NF model. However, experience has shown that that approach leads to relatively long running queries that impact OLTP performance. This lead to the data warehousing concept and (later) dimensional modeling.

Thank you very much for your answer. What about the future with in-memory / vectorial database ?