I am coming off of a failed project where a DW was not built and the data mart development group tried to take over the functionality of a DW and the DW (ie they tried to handle both the intake process and the presentation). The data mart team was using the classic dimensional model. They were dealing with large disparate sources of data from a number of very disparate system (OS, hardware... etc).

My question is this: Is a warehouse in a more normalized form needed to supply (guarantee) success for) a data mart effort, or can purely dimensional structure handle everything? Estimated initial raw data was approximate 3/4 of a terabyte a month.


 

    Requires Free Membership to View

    Login

I see many "once a star, always a star" implementations that are crumbling. Categorically, star schemas cannot do it all! They are an important part of the architecture. The key word being "part". Your intake layer is best served via a relational model that is highly normalized and strikes the right balance between your enterprise data model and your inventory of source application data structures. The intake layer also is designed to capture and retain audit trail history. This is called the PSA. (Persistent Storage Area). The data marts are sourced from the PSA. Data marts DO NOT mean star schemas. Data marts are designed for performance and/or specific information access requirements. Star schemas are very good for those business requirements that need a lot of dimensional analysis and trending analysis. Relational schemas are good for business requirements that need a lot of selective reporting and list management.

 

For More Information


 

This was first published in March 2002

Join the conversationComment

Share
Comments

    Results

    Contribute to the conversation

    All fields are required. Comments will appear at the bottom of the article.
    Back to top