What is the difference between Relational and Multidimensional database implementation ?

This was one of my second question on linked in to get a more industrial perspective answer. A lot of discussion on the implementation of relational or multidimensional database implementation. Some of the feedback from experts in this fields are outlined below :

Michel Voogd :
“The difference in implementation is that a multidimensional database includes pre-packaging subsets of data into small objects that are usable for fast online browsing, usually in a BI portal environment such as Cognos or Business Objects.
A relational database in itself doesn't include those packages but it would allow querying larger datasets.”

Bala Seetharaman :
“Relational DB - ER Modeling it has to comply the Codd's 12 rules. Here you can store only the way supported by DB engine (you can partition or multifile groups)
Multidimensional DB - Dimensional model - store the pre-aggregated data in the multidimensional form, still data sourced from Relational DB or Flat files. (Here you can store in the form of MOLAP, ROLAP and HOLAP and DOLAP too).

SQL - Query language used to search and manipulate the data from Relational DB
MDX - Multidimensional Query Expression - used to search and retrieve the data from cube or MDB (Multidimensional) store.

Siddharth: To answer your question "are there is any different tool or language to query the multidimensional database ( CUBE )",
MDX is the query language used to query the cube like your SQL again, its not like your ANSI standard SQL, we need to write in the form of 3D axis.

the calculations are quite easy in RDB than MDB, here if don't understand the dimension and hierarchy members we can't get the result easily in cube. “
John McKenna :
“….In relational databases data is organized by tables and columns (tuples) and records are grouped into blocks for storage and access. Querying is performed based upon relational algerbra (SQL). In multi-dimensional database implementations (most no longer exist), data is organized into mulit-dimensional cubes (think multi dimensional arrays), and queried based on a language suitable to navigating cubes (I am not aware of a standard although one may exist). To further muddy the waters you have columnar databases that group column data into blocks (efficient for ROLAP applications where few columns are in the result set, therefore less blocks traversed).

In addition to the database implementations many reporting tools have (OLAP/cube) functionality built in but many of these are not full blown multi-dimensional databases but scaled bown persistance engines that store all cube values together. Most full blown multi-dimensional databases have faded away due to performance issues (due to sparcity issues, etc), learning new query languages, supporting multiple database platforms and people finding that it was relatively easy to implement cubes in relational databases (ROLAP) by using dimensional database design (Ralph Kimball). …….”


The next question was the query methodology to query both type of database implementation, luckily I came to know that Oracle has also implemented multidimensional database architecture called the ESSBASE and for SQL Server its SSAS and SSRS.

Well there is still lot of information on my profile if you may want to have a look. Compiling all the notes is actually a tedious job. I have tried to aggregate some of the valuable comments.

For detailed discussion please follow the link