SQL Server - index on a computed column?

Question

I\'m joining to a table dozens of different times, and every time, I join (or filter) based on the results of a SUBSTRING of one of the columns (it\'s a string, but left-pad

Answer 1

Assuming you have your fields in this format:

00Data0007
000000Data0011
0000Data0015

, you can do the following:

• Create a computed column: ndata AS RIGHT(REVERSE(data), LEN(data) - 4)

  This will transform your columns into the following:

  ataD00
  ataD000000
  ataD0000
  

• Create an index on that column

• Issue this query to search for the string Data:

  SELECT  *
  FROM    mytable
  WHERE   ndata LIKE N'ataD%'
          AND SUBSTRING(ndata, LEN(N'ataD') + 1, LEN(ndata)) = REPLICATE('0', LEN(ndata) - LEN('ataD'))
  

  The first condition will use an index for coarse filtering.

  The second will make sure that all leading characters (that became the trailing characters in the computed column) are nothing but zeros.

See this entry in my blog for performance detail:

• SQL Server: leading wildcard match using an index

Update

If you just want an index on SUBSTRING without changing your schema, creating a view is an option.

CREATE VIEW v_substring75
WITH SCHEMABINDING
AS
SELECT  s.id, s.data, SUBSTRING(data, 7, 5) AS substring75
FROM    mytable

CREATE UNIQUE CLUSTERED INDEX UX_substring75_substring_id ON (substring75, id)

SELECT  id, data
FROM    v_substring75
WHERE   substring75 = '12345'