Physical Data Layout
Another way to improve performance is to store each
relational table with a row- or column-oriented physical
layout. No matter which orientation is used, the rows of
each table are spread evenly among the AMPs. But, on
each AMP, either the values within each row are stored
together, or the values within each column are stored
together. Columnar orientation provides extreme performance when a table has many columns, and only a
small number of those columns are generally required
for a query. In this case, the database can read only data
from the required columns off of the disk, reducing the
I/O versus reading all of the columns in a row-oriented
storage layout. Row orientation is more efficient when a
high percentage of the columns are generally required
for queries because there is little excess I/O required
to read unnecessary data, and there is no overhead to
reassemble the rows to be returned to the user. Teradata
can break the table up into any combination of row- and
column-oriented pieces for physical storage within its
parallel architecture. This rabase.