SQL: Speed Up SELECT DISTINCT Queries

by Neil Boyle

Many people use the DISTINCT option in a SELECT statement to filter out duplicate results from a query's output. Take this simple PUBS database query as an example:

SELECT DISTINCT
au_fname,
au_lname
FROM authors

In a simple SELECT from one table (like the one above) this is the easiest and quickest way of doing things.

However, with a more complex query you should think about re-coding it to gain a performance advantage. Take this query for example, which only returns authors that have a book already published.

SELECT DISTINCT
au_fname,
au_lname
FROM authors a JOIN titleAuthor t
ON t.au_id = a.au_id

Here, we only want to see unique names of authors who have written books. The query will work as required, but we can get a small performance improvement if we write it like this:

SELECT au_fname,
au_lname
FROM authors a
WHERE EXISTS (
SELECT *
FROM titleAuthor t
WHERE t.au_id = a.au_id
)

The reason the second example runs slightly quicker is that the EXISTS clause will cause a name to be returned when the first book is found, and no further books for that author will be considered (we already have the author’s name, and we only want to see it once)

On the other hand, the DISTINCT query returns one copy of the author’s name for each book the author has worked on, and the list of authors generated subsequently needs to be examined for duplicates to satisfy the DISTINCT clause.

You can examine the execution plan for each query to see where the performance improvements come from. For example, in SQL 6.5 you will normally see a step involving a Worktable mentioned for the "DISTINCT" version, which does not happen in the EXISTS version. In SQL Server 7.0 and 2000 you can generate a graphical execution plan for the two queries and more easily compare them.

The performance improvement you get depends on the ratio of matching rows in the left and right (or inner and outer) tables. The query below will work in any SQL Server database. Try pasting the two queries into ISQL or Query Analyzer and comparing the execution plan and I/O costs the two produce in different databases. The second query usually comes out as more efficient, though the actual performance gain varies.

SELECT DISTINCT o.name
FROM sysobjects o
JOIN sysindexes i
ON o.id = i.id
WHERE o.type = 'U'

SELECT o.name
FROM sysobjects o
WHERE o.type = 'U'
AND EXISTS (
SELECT 1
FROM sysindexes i
WHERE o.id = i.id
)

You need to understand the relationship between the two (or more) tables you are joining in order to execute this trick properly. The two Northwind database queries below are designed to return customer IDs where a discount of more than 2 percent has been given on any item. At first glance, these two queries appear to produce the same results because they follow the format in the examples above, but the results you get are actually different.

SELECT DISTINCT customerID
FROM orders o
JOIN [order details] od
ON o.OrderID = od.OrderID
WHERE discount > 0.02

SELECT customerID
FROM orders o
WHERE EXISTS (
SELECT *
FROM [order details] od
WHERE o.OrderID = od.OrderID
AND discount > 0.02
)

These examples do not match up because it is OrderID that defines the relationship between the two tables, not the customer name. The second query will return multiple customer names, one for each order placed by the customer. Try adding the OrderID column into the SELECT list to see this.
So the next time you find yourself using the SELECT DISTINCT statement, take a moment to see if it can be re-written for improved performance. You may be surprised at what a little re-coding can do for your application.