The Data Warrior

Is the Data Vault too complex?

This was a very interesting topic that came up on LinkedIn the other day, so I wanted to address it here to.

There seems to be quite a few people who think that Data Vault models are harder to build and manage than what we do today in most shops. So let me explain how I came to learn Data Vault Data Modeling.

Before learning Data Vault, I had successfully built several 3NF OLTP, 3NF DW, and Kimball-style Dimensional data warehouses (and wrote about it with Bill Inmon and Len Silverston in the original Data Model Resource Book).

In other words, I had a reasonable amount of experience in the subject area (data modeling and data warehousing).

I personally found Data Vault extremely easy to learn as a modeling technique (once I took the time to study it a bit). At the time that meant reading the old white papers, attending some lunch & learns with Dan Linstedt and then building a few sample models.

I was definitely skeptical at first (and asked lots of questions at the public lunch & learns). I did not care about MPP, scalability, or many of the other benefits Dan mentioned. I just knew from experience there were a few issues I had seen with the other approaches when it came to building a historical enterprise data store and was hoping Data Vault might be a solution.

In comparison to trying to learn how to design and load a Type 2 slowly changing dimension, Data Vault was a piece of cake (for me anyway).

Once I was convinced, I then introduced the technique to my team in Denver – who had virtually no data warehouse experience.

It was universal – everyone from the modelers to the dbas to the ETL programmers found the technique very easy to learn.

Our investment: One week of training from Dan for 7 people and 3 or 4 days of follow-on consulting where Dan came in once a month (for a day) to do a QA review on our models and load routines and mentor us on any issues we were having.

Dan did not make much $$ off of us. 😦

Since then, I have found that experienced 3NF modelers pick up the technique in no-time flat.

Why is that?

Because Data Vault relies on solid relational principles, experienced 3NF modelers seem to grasp it pretty fast.

Modelers who only have experience with star schemas, on the other hand, seem to have a bit of a hard time with the approach. For some of them it is a paradigm shift in modeling technique (i.e., feels very unfamiliar – “too many tables and joins”),  for others it is almost a dogmatic objection as they were (sadly) taught that dimensional/star was the only “right” way to do data warehousing.

They are just not open to a new approach for any reason (sad but true). 😦

The biggest issue I have seen with clients is a reluctance to try the approach for fear of failure because they don’t personally know anyone (other than me) who has used the approach and because they think it is easier (and cheaper?) to find dimensional modelers.

This happens, even if they agree in concept that Data Vault sounds like a very valid and flexible modeling approach.

As we all know, it takes $$ to train people on star schema design too, so my advice is that if you have a team of people who know 3NF but don’t know dimensional, train them on Data Vault to build your EDW, then hire one or two dimensional modelers to build your end user reporting layer (i.e., data marts) off the Data Vault.

So that’s my 25 cent testimonial. (You get if for free!)

If you want to learn more about Data Vault, check out my presentations on SlideShare or click on the Super Charge book cover (below my picture in the sidebar) to buy the Data Vault modeling book..

Check it out and let me know what you think in the comments. How do we get people over the fear of trying Data Vault?

Talk to you later.

Kent

Data Vault Certification Class – Day 3 (It’a a wrap!)

Well, not so cold today in Montreal. Instead we got very cold rain and snow mix. Yuk. (But I definitely want to come back in the summer!)

This morning, Dan dived into how to load the Data Vault with all new material he has not taught in the certification class before. We really got lucky by attending this class. I knew most of the concepts, and have implemented most of them, but his new slides are just killer. They really get the point across and cleared up a few points for me too.

Not only do the slides include sample SQL for various parts of the load, and the logical patterns, Dan even demonstrated some working code on his really cool, high-powered laptop. It was great for the class to see the Data Vault concepts put into practice. (And he of course had some more tales to tell)

Cool Phrase for the Day

Short Circuit Boolean Evaluation: A mathematical concept, that Dan laid on us, that is used to get very fast results from the Data Vault change detection code . We use it in doing column compares on attribute data to determine if a Satellite row has changes.

In Oracle it looks like this: decode(src_att1, trg_att1,0,1) = 1

In ANSI-SQL it is a bit longer but has the same effect:

CASE WHEN (src_att1 is null and trg_att1 is null or src_att1 = trg_att1))

THEN 1 else 0 = 0

I have been using this for years (learned it from Dan) but had no idea there was a math term for it.

Okay so I am a geek. 🙂

The Test

After all that cool stuff came the certification test.

Not easy. My hand cramped writing out the answers.

We get our results next week. (Dan has a fun weekend ahead of him doing a bunch of scoring).

I am sure everyone in class will do fine. As I said, they all seemed to get it.

Anyway, the class is over now and I am in a hotel in Vermont (where it is snowing now). I fly back to Houston in the morning.

I had a good week here in the northeast (despite the weather). It was definitely worth the time and money to come for this class. I met some great people, learned a lot, and got to spend time with my good friend Dan.

Watch out Montreal – you are about to be descended upon by a whole new batch of Data Vault experts.

It could change the way you do data warehousing.

Later.

Kent