Recently, I have been working on Query Optimization project. While working on it, I found the following interesting observation. This entire concept may appear very simple, but if you are working in the area of query optimization and server tuning, you will find such useful hints.
Before we start, let us understand the difference between Seek Predicate and Predicate. Seek Predicate is the operation that describes the b-tree portion of the Seek. Predicate is the operation that describes the additional filter using non-key columns. Based on the description, it is very clear that Seek Predicate is better than Predicate as it searches indexes whereas in Predicate, the search is on non-key columns – which implies that the search is on the data in page files itself.
Earlier, I had posted regarding how to remove key lookup and bookmark lookup in the following:
SQL SERVER – Query Optimization – Remove Bookmark Lookup – Remove RID Lookup – Remove Key Lookup
Let us see an example where we will remove the bookmark lookup first using the covering index. On removing the bookmark lookup, it resulted in Index Scan, which is not good for performance. When Index Scan is converted to Index Seek, it provides a significant improvement in performance.
Run following SELECT, which is based on the database AdventureWorks.
USE AdventureWorks
GO
-- CTRL + M Enforces Key Lookup
-- Try 1
SELECT NationalIDNumber, HireDate, MaritalStatus
FROM HumanResources.Employee
WHERE NationalIDNumber = 14417807
GO
Let us check the execution plan for the same. The execution plan consists of the key lookup because there are columns that we are trying to retrieve in SELECT as well in WHERE clause.
Let us create a covering index on this table HumanResources.Employee.
-- Create Non clustered Index
CREATE NONCLUSTERED INDEX [IX_HumanResources_Employee_Example] ON HumanResources.Employee
(
NationalIDNumber ASC, HireDate, MaritalStatus
) ON [PRIMARY]
GO
After creating above index, let us run the same SELECT statement again.
-- CTRL + M Removes Key Lookup, but it still enforces Index Scan
-- Try 2
SELECT NationalIDNumber, HireDate, MaritalStatus
FROM HumanResources.Employee
WHERE NationalIDNumber = 14417807
GO
Due to non-clustered index, Key Lookup is removed along with Nested Loops but Index Scan still remains, and this is not good for performance.
I tried to remove the scan, but my attempts were unsuccessful. I finally looked at the datatype of the NationalIDNumber. All this time, I was assuming that this datatype is INT, but on a careful check, I found that the datatype of NationalIDNumber is nvarchar(15).
In the SELECT statement, we were comparing the datatype of NVARCHAR to INT, and this was forcing the predicate operation while executing the query.
As discussed earlier, due to predicate operation, there has to be explicit conversion on the side of NationalIDNumber, which forces the query optimizer to not use the index and instead it has to scan the complete data in table to get the necessary data. This is not the desired solution. Index Scan reduces performance. The reason for this conversion is because I am using INT value in WHERE clause instead of NVARCHAR.
I changed my WHERE clause and passed STRING as the parameter instead of INT.
-- CTRL + M Removes Key Lookup and it enforces Index Seek
-- Try 3
SELECT NationalIDNumber, HireDate, MaritalStatus
FROM HumanResources.Employee
WHERE NationalIDNumber = '14417807'
GO
After running the query with the changed WHERE clause, the Index Scan is now converted into Index Seek.
Index Seek is definitely the most optimal solution in this particular scenario.
When the detail execution plan was checked, I found the following two notable points. First, the predicate is converted to seek predicate, which is the reason for performance improvement, as described earlier. Instead of scanning data in the table, Index Seek is performed. Second, as the datatype of the NationalIDNumber is NVARCHAR and the parameters are passed as VARCHAR, the conversion happens on the parameters instead of NationalIDNumber column, and this forces Index Scan to Index Seek.
If we pass the parameter as NVARCHAR instead of VARCHAR, the execution plan will remain the same, but CONVERT_IMPLICIT will not be required any more. Let us run the following query, which has NVARCHAR as the parameter.
-- CTRL + M Removes Key Lookup and it enforces Index Seek and no CONVERT_IMPLICIT
-- Try 4
SELECT NationalIDNumber, HireDate, MaritalStatus
FROM HumanResources.Employee
WHERE NationalIDNumber = N'14417807'
GO
The execution plan of the above query is very similar to that in which we had passed the parameter as VARCHAR.
Now let us check the execution plan for the same.
In the WHERE condition, the operators we have on both the sides of “=” are of NVARCHAR. NationalIDNumber and parameter passed – both are NVARCHAR, which has removed CONVERT_IMPLICIT. However, there are no changes in the execution plan.
In summary, when Key Lookup is removed and index seek replaces index scan, the performance is tuned up. Let us quickly compare the execution plan of the above four options. I have included the complete code here for easy reference.
USE AdventureWorks
GO
-- CTRL + M Enforces Key Lookup
-- Try 1
SELECT NationalIDNumber, HireDate, MaritalStatus
FROM HumanResources.Employee
WHERE NationalIDNumber = 14417807
GO
-- Create Non clustered Index
CREATE NONCLUSTERED INDEX [IX_HumanResources_Employee_Example] ON HumanResources.Employee
(
NationalIDNumber ASC, HireDate, MaritalStatus
) ON [PRIMARY]
GO
--WAITFOR DELAY '00:00:30'
-- CTRL + M Removes Key Lookup, but it still enforces Index Scan
-- Try 2
SELECT NationalIDNumber, HireDate, MaritalStatus
FROM HumanResources.Employee
WHERE NationalIDNumber = 14417807
GO
-- CTRL + M Removes Key Lookup and it enforces Index Seek
-- Try 3
SELECT NationalIDNumber, HireDate, MaritalStatus
FROM HumanResources.Employee
WHERE NationalIDNumber = '14417807'
GO
-- CTRL + M Removes Key Lookup and it enforces Index Seek and no CONVERT_IMPLICIT
-- Try 4
SELECT NationalIDNumber, HireDate, MaritalStatus
FROM HumanResources.Employee
WHERE NationalIDNumber = N'14417807'
GO
/* What is the reason for difference between Try 2 and Try 3?
Check the exeuction plan
*/
-- Drop Index
DROP INDEX [IX_HumanResources_Employee_Example] ON HumanResources.Employee WITH ( ONLINE = OFF )
GO
Let us look at the execution plan.
Summary points:
- In general, Index Seek is better than Index Scan (I am not talking about it depends conditions)
- Understand Predicates and Seek Predicates and see if you have only Seek Predicates.
- In case of key lookup or bookmark lookup, see if you can create a covering index or included column index.
- Use the datatype wisely even though there is no change in the resultset.
Reference: Pinal Dave (http://blog.SQLAuthority.com)
.jpg "Bill, Blythe, Pinal")
.jpg "Veg Maxican Meal")
.jpg "Chief Seattle")
.jpg "SQL PASS Summit 2009")
.jpg "PASS Registration")
.jpg "Community Leads - Joe Webb, Pinal Dave and Arnie Rowland")
.jpg "Pinal Dave and Good Old Friend Steve Jones")
.jpg "Brent Ozar (very left) and Excellet Fellow SQL Enthusiast")
.jpg "Midnight DBAs - (Sean and Jen) and Pinal")
.jpg "Pinal Dave and Michelle Ufford (SQLFool)")
.jpg "SQL Server Regional Mentor Team and Rushabh Mehta")
.jpg "Allen White (with his wife), Jacob Sebastian and Pinal Dave")
.jpg "SQL PASS Summit 2009 and Pinal Dave")
.jpg "PASS Quizbowl")
Rushabh Mehta is a Mentor for Solid Quality Mentors’ global Business Intelligence Division and the Managing Director for Solid Quality India. Rushabh has worked with Microsoft SQL Server since 1998 and has designed and developed BI systems on the Microsoft platform since 2001, including a predictive analysis system, a health-care analysis solution, and a multi-terabyte financial decision support system. He has been the lead architect for many large-scale and complex BI solutions for clients such as Publix, Raymond James Financial, Jackson Hewitt, and Checkers. Rushabh has been an active volunteer with PASS since 2000 and has also served on the board for a number of years. Rushabh is also a SQL Server MVP and a frequent speaker at large conferences and PASS Chapters around the world.
pinaldave
