A New Internet Library: Add Your Website/Blog or Suggest A Website/Blog to our Free Web Directory http://anil.myfunda.net.

Its very simple, free and SEO Friendly.
Submit Now....

First, an up front apology to Mosha - this is half in jest, half real. MDX is a powerful and useful language, but I find that its very existence has made my life very difficult when it comes to developing a solid Analysis Services solution. That's not to say I don't use it, or endorse it, but I worry greatly about its overall usage.

1) It tempts you by looking like SQL, but once your brought into it you realize how complex it is (if you don't have Mosha's MDX Studio, get it).

2) MDX allows you to do TOO much when it comes to a multi-dimensional structure, without first understanding multi-dimensional concepts. I've seen code that creates filters or subset cubes that overlap or end up cutting out measures implicitly. You have no warning about possibly undermining previous dimensional modeling - that would be a nice addition to see.

3) Along those lines, it enables sloppy dimensional modeling in the first place. "Throw it together, will add in things later". With a multi-dimensional database, it is very important to model it correctly, just like you would a relational structure. Unfortunately there are no Codd/Date rules of normalization, and often very little work done up front understanding the actually relationships and what the data is meant to represent. For example, instead of using a many-to-many cube dimension (what I call the dimesion to measure tie) to represent something like a bill of materials, I often see this done afterwards somehow via MDX calculations and such. Instead of going back to the dimensional modeling, the answer is usually "we can do this in MDX later".

4) Yet another language with its own syntactical oddities - damn you syntax, you will be the death of me!

5) I can't count the number of calculations I've seen in a cube that could have been done before the cube is loaded in the first place. An MDX calculation that does something like PRICE * QUANTITY = TOTAL SALES is just a waste of processing on the cube side. Remember, MDX is post aggregation - so the large the cube, the more dimensions it has and the level of drill down it allows all play a roll in the speed of things.

Those are 5 to start - anyone have anymore?




Source Click Here.

0 comments