A columnar DBMS is an implementation of the relational theory, but with a twist. The data storage layer does not contain records. It contains a grouping of columns.
Due to the variable column lengths within a row, a small column with low cardinality, or variability of values, may reside completely within one block while another column with high cardinality and longer length may take a thousand blocks. In columnar, all the same data - your data - is there. It's just organized differently (automatically, by the DBMS).
The main reason why you would want to utilize a columnar approach is simply to speed up the native performance of analytic queries. There is no difference in the SQL or data access tool used to interface with the data or the logical data modeling.
The columnar approach to data management is not new. In fact, the mainframe database Model 204, first deployed in 1972, used a similar approach. However, columnar's time may have finally come as analytic processing becomes a business imperative.
Learn about the columnar orientation and how it can be effective for your needs. This is the orientation of Vertica, ParAccel, Sybase IQ, InfoBright, Exasol, VectorWise and MonetDB. As well, Microsoft SQL Server, in its 2010 release, will have an optional column-oriented storage layer.