Archives of the TeradataForum
Message Posted: Tue, 20 Mar 2001 @ 22:10:46 GMT
| Subj: || || Re: Why not use star schema |
| From: || || James Judge |
Obviously since this is a Teradata Forum you have heard all the reasons for a "centralized" data image versus "the other" warehouse
methodologies (data-hub, data-marts, staging area, etc.). Even Kimball, et. al., state that operational schemas are (should be) normalized.
The issue is then how much of the normalized detail data should be in the data warehouse.
As you know Teradata's position is, detailed data with as much history as is cost effective. I think what happens at that point is from
an ETL perspective (i.e. how easy/hard is it to capture the "events" from the operational systems). A lot of operational too suites have
their data stores as "state engines" , meaning they only retain the current state of the data and not necessarily the events that caused the
There can be all kinds of reasons (political, costs, scope of solution) that then determine what data is sourced. If only "state data"
is available then perhaps a "dimensional model" would be better (but being a Teradata bigot I can't believe that). We can still load "state
data" in it's atomic form and build (logical or physical) dimensional views on top of that, in one Teradata instance (and I don't see the
DBMS's taking that approach in implementation.