The Atomic Data Warehouse
Posted at 11/12/2008 09:16:00 AM
No, this isn’t a post about a place to store information on nuclear weapons. The long-running debate between Bill Inmon and Ralph Kimball over the optimum data warehouse model has come up again in a discussion of our approach to data warehouse structure.
So, should it be a bus (Kimball) or a CIF (Inmon)? Both of these models recognize the importance of having atomic-level data available and both recognize the limitations of the relational model in providing access to this data. They propose two different approaches that are intended to provide the maximum benefit to the business within the constraints of the relational model.
“Within the constraints of the relational model”! What if there was an option that did not have those constraints? I think it is time to speak in complete sentences. Rather than saying, “Model x is the best approach”, one should say, “Model x is the best approach when restricting the options to relational databases.”
OK, end of rant.
There really are other viable options, even for large EDWs. The correlation database model (CDBMS) provides a single level, centrally managed structure that still gives users the option to have their own data mart. Using the incremental query provided by illuminate, the first query step creates the logical data mart. Once that is done, the user can analyze the results just as they would with a free-standing data mart. However, they would be using managed, current and consistent data rather than the chaotic departmental data marts that are seen today in the RDBMS world.
So, should it be a bus (Kimball) or a CIF (Inmon)? Both of these models recognize the importance of having atomic-level data available and both recognize the limitations of the relational model in providing access to this data. They propose two different approaches that are intended to provide the maximum benefit to the business within the constraints of the relational model.“Within the constraints of the relational model”! What if there was an option that did not have those constraints? I think it is time to speak in complete sentences. Rather than saying, “Model x is the best approach”, one should say, “Model x is the best approach when restricting the options to relational databases.”
OK, end of rant.
There really are other viable options, even for large EDWs. The correlation database model (CDBMS) provides a single level, centrally managed structure that still gives users the option to have their own data mart. Using the incremental query provided by illuminate, the first query step creates the logical data mart. Once that is done, the user can analyze the results just as they would with a free-standing data mart. However, they would be using managed, current and consistent data rather than the chaotic departmental data marts that are seen today in the RDBMS world.
