Column Partitions With ROW Format
A physical row of a column partition with ROW format is called a subrow. Subrows have the same format as traditional Teradata Database rows, but only include the columns defined for the column partition. Note that currently there is no autocompression for subrows.
If Teradata Database uses ROW format for the column partitions of a column‑partitioned object is such that many narrow column partition values occur, there can be a very large increase in the space needed for a column‑partitioned object compared to the object without column partitioning. In the worst case, the space required for the column‑partitioned object can be as much as 24 times larger.
Because of this, you should not specify ROW format for narrow column partitions. In the worst case, each column partition value can have a row header (14 or 20 bytes plus 6 or more bytes needed for a container) for each column partition value as compared to the same object without column partitioning. An object that is not column‑partitioned only has a row header (14 or 20 bytes) for each regular row.
ROW format is useful for wide column partitions where one or only a few values fit in a container, and there is neither much benefit nor a negative impact from row header compression. ROW format provides quicker and more direct access to a specific column partition value than with COLUMN format because when a container has COLUMN format, Teradata Database must locate the column partition value within the container.
Depending on the autocompression types used for a container, access can be as simple as indexing into the container or it might require a sequential access through bits indicating how a value is compressed or sequential access through the column partition values, or both, to position to the specific column partition value to be accessed.
See “Row Structure for Subrows (ROW Format)” on page 767 for more information.