Hi All,
We all are talking lots about Dimensions and fact table in
warehouse. I would like to know few thing from experts:
1) Dimensions and facts are in Warehouse or Datamarts.
2) If we have Dimension and facts in Warehouse and datamart then what
the difference in design.

Waititng for reply.

Regards
Umesh Choudhary

Not really a guru in DW construction .... but my 2c worth ...

Datamarts are really just a subset of the DataWarehouse. The design
principles are the same, but focus is on just one Geography or one
Business Area (however you choose to organise your business).

If geographic, then the DataMart may reflect most of the structure of
the Data Warehouse but with less populated tables (however the
Dimension tables might still be fully populated). A Business-Area
centric DataMart may contain all data for only a few of the many
related conbinations of Fact & Dimension tables.

Essentially no difference in design -merely a matter of scope.

Michael


On 28 Aug 2004 00:43:33 -0700, umesh (AT) apnaloan (DOT) com (Umesh Choudhary)
wrote:

<YouveGotToBeKidding (AT) nowhere (DOT) com> wrote

Data warehouses are assumed to have a longer life and contain finer
detail, more likely to be in a normalised form. They will almost always
be relational and will be structured for maintainability rather than
query performance, so they wouldn't normally need to include aggregates.

Data marts are focused on supporting a particular set of queries
directly and are more likely to contain aggregates. They may be
relational or multidimensional, the latter being favoured if the
calculations are complex and fast performance is a must. If
multidimensional, they will have a completely different structure to the
enterprise data warehouse. Even if relational, they will probably have
fewer dimensions and less detail.

Comments at bottom:

On Sat, 28 Aug 2004 10:35:05 +0100, "Nigel Pendse"
<nigelp.nospam (AT) compuserve (DOT) com> wrote:

An interesting conversation evolves from a simple question....

The Data Warehouse guys I've worked with follow Kimbal. They struggle
with the heavily-normalised Data Warehouse concept that is normally a
Teradata agenda.

We observe that Relational DataMarts (ie excluding MOLAP cubes) are
falling out of fashion in favour of accessing a singular central
Warehouse. Historic concerns about Bandwidth and Performance appear to
be fading as technology advances. The warehouse is then best
structured as per Kimbal to satisfy Business Users Queries.
Before that we'd then have a very normalised Operational Data Store
("ODS") or Staging Database -but its performance is generally
unsuitable for Client Access. It seems suitable only as an ETL data
source for population of the Multidimensional Warehouse.

I guess we differ only semantically. If I understand correctly: Where
you have a Normalised DataWarehouse and Dimensional DataMarts, I'd
have a Normalised ODS then Dimensional Warehouse, then Datamarts
containing only a subset of the Warehouse Data.

Michael
(michaelr at usa.net)