Database Reference
In-Depth Information
terms of not having to scan over all the attributes of a row when scanning
for large aggregations. It's a big benefit when a lot of attributes are present,
as in a typical data warehouse scenario.
Figures 3-3 and 3-4 show the differences between a column store and a row
store.
FIguRe 3-3
How a row store selects data
FIguRe 3-4
How a column store selects data
An additional benefit of column store databases is the compression that's
achievable. As similar data is stored contiguously, both dictionary compres-
sion and run length encoding are successful techniques. Read more on this
at Daniel Abadi's excellent paper, which you can find at
http://db.csail
.mit.edu/projects/cstore/abadi-sigmod08.pdf
.
Another key benefit of column store databases (as proven in the Abadi's paper,
as well as in much other research) is that denormalization is not as necessary
for performance reasons in a column store. This is the reason the latest ver-
sion of Analysis Services has a “tabular” mode (or at least that it is called that).
