Do I need a Data Vault

I want an ODS that feeds a Dimensional Model. The ODS should have all history from the transaction system with little or no transformation.

But some suggest scrapping the ODS and building a Data Vault in 3NF that feeds the Dimensional model. The Data Vault is billed as a Hybrid to Immom. It seems to me that it's a way for 3NF people to create their 3NF Data Warehouse without having to do everything that Immom says you need to do.

Thoughts?

If it ain't broke, don't fix it. I am intrigued by your desire to store all history twice. That seems a little redundant. :-)

Those 3NF folks aren't so bad. It occasionally still pays the bills.

The purpose of an ODS it to provide operational query support, not a historical data repository.  It is usually a subset of data to support specific applications, such as a customer support desk.  It never has full history, usually just enough to cover the time periods necessary to support the intended applications.

When you talk about Lindstedt's Data Vault or Inmon's architecture, you are talking about a store and publish architecture where all data is stored in a central repository (Data Vault or a 3NF model) which is not directly accessed by end users.  Instead a publication function exists to pull the data from the repository for analytic use (dimensional marts, file extracts, cubes, etc…).  These are not, in any way, shape, or form, an ODS.

Kimball's architecture is a store and use architecture.  Users have direct access to the dimensional repository for analytics.  It is just as valid as Lindstedt's and Inmon's approaches and equally capable of maintaining an enterprise wide data repository.

All of these architectures view an ODS as an ancillary repository to address specific use cases.  There are many successful implementations of all of these architectures.  Choose your poison...

A couple of things......

ODS seems to have a couple of different definitions. One is that it has very little history and is used, I think, primarily to support Real Time Reporting. The other definition that I have seen is the ODS has lots and lots of history in databases that look a lot like the transaction systems and has very little transformation.

If the source systems are feeding a Data Warehouse that has all the data, then certainly an ODS with little data is all that is needed. But, if the data warehouse does not have all the data from the transaction systems, then having a place that keeps all the history in it's native form might be a good thing. Data can be deleted from the transaction system and remain in the ODS. This data can be loaded into a DW without having to reload old data from backups.

I don't mind keeping history in 2 places if in one place the data looks like the transaction system and in the other place, it is in a dimensional model.

Immon's Data Warehouse or a Data Vault with a presentation area in a dimensional model seems like the type of history in 2 places that is unnecessary. I see value in Immom and a Data Vault, but I'm not sure the value outweigh's the cost. I'm sure there are instances when the value does out wiegh the costs and times when it doesn't. Otherwise, everyone would either have a formal Staging Area in 3NF or no one would.

I would love to hear when one approach makes more sense than the other approach.