​
Indexing nullable columns

When a nullable column is indexed, the NULL value is considered to be lower than any value of the type of the column. This means that it is possible to have an index segment with only the NULL values by specifying a first range separator that is smaller than any expected value of that column. (Note: In the case of a string type, the minimal possible value is the empty string ''.)

​
Indexing high frequency values

Sometimes, in the case of a non-unique index, the indexed column has a few values that appear a large number of times. It is possible, using range separators, to index each such value in an index segment of their own. This has an advantage that values that collate close to those values will not be in large index segments.

As an (artificial) example, consider an orders table that has an order_status column.
Suppose that there is a large number of possible values that order_status could have.
Also, there is a large number of rows with status delivered and returned.
There are also an unknown number of orders that don’t have a status, that is, their status is NULL. Without getting into too much business logic in this artificial example, suppose that scans are likely over order_status values that are not delivered or returned. For example, periodically the application may want to send reminders for orders that are unpaid. The following definition of the range separators would put the index for rows with order_status of NULL, delivered and returned into separate index segments that contain only those values:

CREATE TABLE orders (

	…,

	order_status VARCHAR(11) INDEX WITH (

		DEVICES = (m3d0, m3d1, m3d2, m4d0, m4d1, m4d2),

		RANGE_SEPARATORS = ('', 'delivered', 'delivered ', 'returned', 'returned ')),

	…

);

Note that this example has 6 devices for the 5 range separators, as required. The index segment on the first device holds the index for all the values strictly less than the first range separator. In this example, the first range separator is the empty string, ''. As explained in above, all the rows that have a NULL order_status will be indexed in this index segment (and only those rows).

There is a subtle difference between the range separators 'delivered' and 'delivered '. The difference is the extra space at the end of the second. The only strings that are greater than or equal to the first but less than the second are those that are exactly 'delivered', so only rows with this value as their order_status will be indexed in this index segment. The same logic holds for 'returned' and 'returned '.