>> Data warehouse (DW) modeling (IMO) is more "physical" than logical; the "stars" you develop are usually implemented that way physically too.
>
>OK. But then what sort of methods do you use at the logical modelling stage? I'm a bit of a nut about working through the logical design first. I like to get the ideal world identified before invariably chopping away at the edges (performance, cost and existing infrastructure issues have that annoying habit of creeping in) to produce a desirable physical implementation plan.

I guess I don't understand the "question".

Any "logical" modelling to be done, I would do at the "corporate" model design stage, using whatever "classical" techniques I felt comfortable with. This model would satisfy the "operational" needs of the company and would automatically be the ideal solution for OLTP (eg. normalized).

At the DW modelling stage, I take the "corporate" model and remove any purely operational data, create "time/revision" elements, "denormalize", group elements as to volatility, etc... Whatever "I" felt like, based on the capabilities of the OLAP DBMS I had to work with.

All these last activities are performed for the sake of efficicient OLAP processing (loading and inquiry). And I would only consider the specific subject areas that are of interest. At least, that's the way I see/do it (right or wrong).

P.S. Some consider "snowflakes" a "normalized" extension of "stars" and that taking snowflakes to an extreme will ultimately lead one (back) to the corporate data model; ie. one better suited for OLTP than OLAP.