Speed Tips for Select Statements

You can code a SQL select statement in a number of ways to get
the same results, but some versions of the same query may perform better than others. In this article we will look at ways to exploit this.

Here is a query that I cut-and-paste straight out of SQL 7
Books Online. The query runs in the Northwind database and is designed to pull
out the maximum unit price for each order in the database.

Click here for code example 1.

This type of query is called a Correllated Subquery–you
can see that there are two Select statements, the ‘main’ one (SELECT
Ord.OrderID, Ord.OrderDate ) which selects the order ID and date from the orders
table, and then the ‘sub’ query (highlighted), which
selects the maximum unit price for each order.

The ‘sub’ query is actually ran once for each row that the
main query returns, and this repeated access to the [order details] table can be
fairly inefficient.

Books Online goes on to say that queries like the one above
can usually be re-written as a straightforward join–the example below uses an
inner join between the Orders and [Order Details] table in association with the
MAX() function to produce exactly the same data, but more efficiently.

Click here for code example 2.

Although the same data is returned by both queries, Query
Analyzer indicated that the second
version took around 40% less SQL Server resources to run than the first, so no
prizes for guessing which is the preferred option. However, in some cases we can
use a third method to gain an even greater performance improvement.

Derived tables

Using a derived table is in effect like using a temporary
table, without the hassle of specifically creating and referring to one. I have
re-coded the BoL query to use derived tables in the example below:

Click here for code example 3.

The highlighted code causes the SQL Server optimiser to generate a
notional (or derived) table called OrdDet for the duration of the query. The
derived table notionally takes up much less space than the original [order
details] table, because it contains only two columns and only one detail row for
each order. Because of this, my ‘derived table’ version of the query should
run even faster than the Join version. When I checked the
execution plan for the derived table version against that of the
“join” version to see what sort of improvement I got, the results came
out… exactly the same!


I got no improvement there at all! Both queries generate the
same execution plan, and use the same amount of SQL Server’s resources to
return the data. So is my theory all blown to hell? Not quite…

The key to understanding why the derived table technique may
or may not
produce a more efficient result is understanding the query optimiser.

The query optimiser looks at all SQL queries and works out the
most efficient way of accessing the tables used in the query, primarily by using
index statistics. While by re-coding my query I have given SQL Server a
different set of instructions for how to get at the data I want, SQL Server has
decided in both cases that the same method – or execution plan–is the
optimal one. This is always not the case though.

Following exactly the same principle, here is a Group By and Derived table query that produce different execution plans to return exactly the same data. Again, both are for
the Northwind database:

select companyName,
  from   orders o inner join customers c
  on     o.customerID = c.customerID
  group  by companyName
  select companyName,
  from   customers c inner join  (
         select customerID, 
                max(orderDate)as orderDate 
         from   orders
         group  by customerID) as o
  on     o.customerID = c.customerID
  group  by companyName

This time the optimiser chose to use different execution
plans for the two queries, and the derived table version of the query comes up
with roughly 30% improvement in terms of resources used to run the query.

Wrapping Up

You can see from the examples that the Query Optimiser
sometimes needs a little help in picking the most efficient way to execute a
query. It’s worth coding up a couple of versions of critical queries and
comparing their performance characteristics to find the most efficient way of
doing things.

Further reading

Neil Boyle
Neil Boyle
Neil Boyle left school at the age of sixteen thinking that computers were things that only existed in Star Trek. After failed careers as a Diesel Mechanic, Industrial Cleaner, Barman and Bulldozer Driver he went back to college to complete his education. Since graduating from North Staffs Poly he has worked up through the ranks from Trainee COBOL Programmer to SQL Server Consultant, a role in which he has specialised for the past seven years.

Latest Articles