Adherence to dimensional modeling in practice

I prefer following dimensional modeling principles especially those proposed by Kimball and most of the techniques discussed in this forum, I hear a lot of comments from colleagues and friends that dimensional principles are more bookish and no one follows them in day to day modeling. I have observed this kind of conclusions from professionals with ETL and report designing background trying their hands in modeling or doing the job in the absence of a fulltime modeler.
I would like to hear opinions from experienced dimensional data modelers who have seen or been part of substantial number of data warehouses.
What does your experience conclude? How true is the statement that most of the dimensional modeling principles are not followed in practice?

Dimensional modeling works just fine, thank you. The thing is a lot of data modelers don't 'get' it. Its a different way of looking at data and how you deal with it. A lot of traditional data modelers simply can't break old habits.

Likewise, there are a lot of datawarehouse developers that don't get it. Fact tables with 300 columns, all denormalized from the dimensions since "joins are bad". Most companies don't hire qualified data modelers and they end up with hybrid solutions from developers who haven't bothered to read the book.

Yes. I have seen many people confusing dimensional modeling with denormalization. While the right approach is to denormalize the dimension tables and most often normalize the fact table.

i think a lot of the differences come down to experience as well.

I've observed modellers/designers who start out with good dimensional models, but then as typical project "issues" come up (data quality, report requirements etc) the adjustments they make to their solutions tend to not follow best practices - as Ngallemo mentioned, old habits can be hard to break, and when asked for a "quick-fix", often people gravitate to what they know.

Experienced Dimensional modellers have more tools in their arsenal to deal with with the oddities/exceptions that exist in the real world.

BoxesAndLines wrote:Most companies don't hire qualified data modelers and they end up with hybrid solutions from developers who haven't bothered to read the book.

That’s the real problem, people hate spending time reading books and articles written by experts, the only credible source where you can find best practice/solutions. Even some people do have books on their shelves, they only get the half picture and come out with hybrid solutions which sometimes are even worse than the old OLTP solutions. The sad thing is, people tend to resist correcting their mistakes as the stake is too high to accept the failure. Then other people jump to the conclusion that dimensional modelling is bad.

To be a successful data modeller, you are not just skimming through the books, you should try the theory and come back to digest the books to make things really work.

I love the term “Refactoring”, meaning you keep restructuring your work even though it is workable in production. I think the term was born out of OOP bandwagon in application programming, I mean a good bandwagon. I guess refactoring is even more crucial for dimensional modelling as our early attempts are most likely half cooked. Let’s face it, the technology is constantly evolving itself, why can’t we keep our work updated. We also need a bandwagon effect for dimensional modelling, a bandwagon full of success stories, so that people have faith on the technology.