SCD into NDS or DDs

Hi,

I am a student who try to learn Data Warehousing and now I'm doing an intership in an enterprise where I should implement a data warehouse with some application.

I used as data flow architecture, the NDS - DDS data flow architecture. So I ETL all data from source system and I put them into a NDS (Normalized Data Store) and after I ETL data from NDS to DDS (Dimensionnal Data Store).

My problem, now, is that I don't know where I should maitain the data logging. What I mean is where should I indicate and maintain the SCD type of some table.

Thanks in advance.

I don't understand what you mean by 'maintain the SCD type of some table'.

There is no such thing as a SCD in a normalized data model. You may have a table that maintains history, but it is not a SCD. A SCD is a term specific to dimensional modeling, not ER modeling.

If the question is 'where do you keep history?', you must have history in the normalized data store if your intent is to implement a store-and-publish architecture (i.e. Inmon). Publishing history to the dimensional data marts is optional, depending on the needs of the mart.

I know that SCD is a specific term to dimensional modeling.

What I am talking about is how can i do if i would like to have something as SCD in the dimensional modeling into the NDS question of maintening historical changes of the source system ?

Since SCD in the dimensional modeling allow us to keep history, I think there is a way to do the same thing in the NDS. But i don'k know how.
In this entreprise, some transactionnal data undergo some changes so i would like that during ETL into the NDS i would like to save them into the NDS before ETL them in SCD 2 into the dimensionnal data store ..

This is the first time that I look about Store-and-Publish Architecture and also about Publishing history. What does it mean please ?

A store and publish architecture is one where you maintain a permanent normalized data warehouse (store) which is then used to feed data marts (publish). Users perform analytics on the marts, the data warehouse primary function is to store data and serve as a source for the marts. This is the data warehouse architecture proposed by Bill Inmon.

If you are creating a dimensional data warehouse, you do not have a permanent normalized data store. All data is stored in a dimensional model and accessed directly by end-users. This is Kimball's architecture.

As far as keeping tract of historical changes in a 3nf model, it's pretty much the same as in a dimensional model, except you usually include effective date in the PK rather than assigning unique surrogate keys.

I think what i decided to put in a store and publish architecture .... So I ETL data from source system and store them into a permanent normalised data store (NDS). And then I ETL data from NDS to put them into data mart. Is that correct ?

Now into the store and publish architecture, keep change into the source system, about this sentence : "As far as keeping tract of historical changes in a 3nf model, it's pretty much the same as in a dimensional model, except you usually include effective date in the PK rather than assigning unique surrogate keys." I don't understand it correctly. Could you please give me more explanation with an example if possible ?

Thanks in advance.

I wonder whether we meet that criteria (I assume not). But I'm assuming that we'll still see benefits from filegroups in general given our hardware configuration????


_________________
JBWSKS