Cost of Dimensional vs. Relational approaches

[klm2klm2]

There is a theory that the cost of creating, implementing, and maintaining a dimensional DW is less than that of the relational model because the dimensional model is simpler. Does anybody have any knowledge (or even ideas) about how much the dimensional approach saves over the relational approach in terms of development, software, and/or hardware?

Thanks,
Kim

 

[entaroadun]

It depends on how much data and how complicated.

You hit a functionality wall with MOLAP, which is very costly to get past with custom programming. ROLAP is easier to customize, mainly because SQL is a standard, and it's easy to find programmers who are good with it.

However, MOLAP is cheaper in terms of development time for basic stuff, especially in ETL and performance tuning. I don't know of any ROLAP tools that automatically produce the myriad of pre-calculated agg tables that MOLAP has by nature, or any ROLAP engine that can do member-wise selection or drill-down as quickly.

MOLAP's slightly cheaper because they bundle; ROLAP requires a database engine and a reporting engine. The box and disk are basically the same; if you need to customize these, you probably need ROLAP.

 

[johnherman]

Take a look at thread353-907540
The Kimball approach, directly to atomic level dimensional structures saves considerable disk space. The Inmon method assembles all the data in a 3NF relational structure before moving into dimensional data marts. This at least doubles the storage. However, disk is relatively inexpensive.

-------------------------
The trouble with doing something right the first time is that noboby appreciates how difficult it was.
- Steven Wright