A single transaction in SQL Server is the lowest common denominator for performance tuning. When it comes to capturing and analyzing data at this level, the best native tool is SQL Server Profiler. Analyzing the results from SQL Server Profiler enables you to discern a great deal of information about your SQL Server; the collected data reflects the actual processing on the server, not just theory or conjecture.
From a performance-tuning perspective, Profiler gives you the option to filter data and then store it in a table for additional analysis. In a previous tip, SQL Profiler: Features, functions and setup in SQL Server 2005, I outlined the steps in a sequential manner. Here I will focus on how to analyze Profiler results in order to find opportunities for query performance improvements in two common scenarios. The first is based on a single user's issues, and the second establishes key SQL Server query performance issues.
As SQL Server Profiler captures user transactions, the results are stored in a table with a maximum of 41 columns. Some of the columns have Profiler system-related data, but most have valuable raw data that can be queried and analyzed to determine areas for improvement. Before analyzing the two scenarios, the table below outlines a subset of the columns that are key to the data analysis.
| ID |
Column |
Data Type |
Description |
| 1 |
EventClass |
int |
Classification used by Profiler to identify a super class for the row and/or overall transaction |
| 2 |
TextData |
ntext |
T-SQL statements issued |
| 3 |
ApplicationName |
nvarchar(128) |
Registered application in SQL Server |
| 4 |
NTUserName |
nvarchar(128) |
Windows user without the domain |
| 5 |
LoginName |
nvarchar(128) |
The Windows or SQL Server login |
| 6 |
cpu |
int |
CPU usage for the statement or transaction |
| 7 |
Reads |
bigint |
Number of reads performed by the statement or transaction |
| 8 |
Writes |
bigint |
Number of writes performed by the statement or transaction |
| 9 |
Duration |
bigint |
Elapsed time by the statement or transaction |
| 10 |
SPID |
int |
System process identifier in SQL Server |
| 11 |
StartTime |
datetime |
Start time for the statement or transaction |
| 12 |
EndTime |
datetime |
End time for the statement or transaction |
| 13 |
DatabaseID |
int |
Database identifier where the transaction or statement is performed, which is the value from the sys.databases.database_id view |
| 14 |
ObjectID |
int |
Database object identifier for the transaction or statement, which is the value from the sys.all_objects view |
| 15 |
DatabaseName |
nvarchar(128) |
Database name where the transaction or statement is being performed, which is derived from the sys.databases view |
| 16 |
HostName |
nvarchar(128) |
Host name where the transaction or statement is originating |
| 17 |
NTDomainName |
nvarchar(128) |
Domain name where the transaction is originating |
| 18 |
ServerName |
nvarchar(128) |
SQL Server name and instance name |
| 19 |
IndexID |
int |
Index identifier used for the table |
| 20 |
RowCounts |
bigint |
Number of rows returned by the statement or transaction |
A common scenario is a user or department of users experiencing an unknown performance issue. Two options are available to capture the needed data for analysis. The first is to filter the data initially captured by Profiler to only capture a specific user's transactions for a finite period of time. The advantage is a small set of data that could be manually reviewed. The disadvantage is that this data would not show how another process impacts the user. The second option is not to limit the data captured via filtering in Profiler. This gives you the ability to capture and subsequently query the data in order to review all data associated with and potentially impacting a user's session.
| CPU |
Duration |
-- Top 10 statements by CPU usage
SELECT TOP 10 *
FROM [ProfilerTest].[dbo].[ProfilerResults]
WHERE LoginName = 'Domain\User'
ORDER BY CPU DESC |
-- Top 10 statements by duration
SELECT TOP 10 *
FROM [ProfilerTest].[dbo].[ProfilerResults]
WHERE LoginName = 'Domain\User'
ORDER BY Duration DESC |
| Reads |
Writes |
-- Top 10 statements by reads
SELECT TOP 10 *
FROM [ProfilerTest].[dbo].[ProfilerResults]
WHERE LoginName = 'Domain\User'
ORDER BY Reads DESC |
-- Top 10 statements by writes
SELECT TOP 10 *
FROM [ProfilerTest].[dbo].[ProfilerResults]
WHERE LoginName = 'Domain\User'
ORDER BY Writes DESC |
| RowCounts |
ObjectID |
-- Top 10 statements by rows returned to the user
SELECT TOP 10 *
FROM [ProfilerTest].[dbo].[ProfilerResults]
WHERE LoginName = 'Domain\User'
ORDER BY RowCounts DESC |
-- Most used objects by the user
SELECT COUNT(ObjectID) AS 'ObjectCount',
OBJECT_NAME(ObjectID) AS 'ObjectName'
FROM [ProfilerTest].[dbo].[ProfilerResults]
WHERE LoginName = 'Domain\User'
GROUP BY ObjectID
ORDER BY ObjectCount DESC |
Note: In the code snippets above, replace the FROM clause, i.e., [ProfilerTest].[dbo].[ProfilerResults] with the database, schema and table where you captured your results. In addition, replace the 'Domain\User' with the associated user's value.
The second analysis is much broader and requires that you determine the queries where potential issues could occur.
| Averages |
Max Values |
-- Average usage for key metrics
SELECT AVG(CPU) AS 'AvgCPU',
AVG(Duration) AS 'AvgDuration',
AVG(Reads) AS 'AvgReads',
AVG(Writes) AS 'AvgWrites',
AVG(RowCounts) AS 'AvgRows'
FROM [ProfilerTest].[dbo].[ProfilerResults] |
-- Highest usage statements
SELECT MAX(CPU) AS 'MAXCPU',
MAX(Duration) AS 'MAXDuration',
MAX(Reads) AS 'MAXReads',
MAX(Writes) AS 'MAXWrites',
MAX(RowCounts) AS 'MAXRows'
FROM [ProfilerTest].[dbo].[ProfilerResults] |
| Specific Host |
DISTINCT Applications |
-- Transactions count by machine
SELECT COUNT(HostName) AS 'HostCount', HostName
FROM [ProfilerTest].[dbo].[ProfilerResults]
GROUP BY HostName
ORDER BY HostName DESC |
-- Determine all applications on the SQL Server
SELECT DISTINCT(ApplicationName)
FROM [ProfilerTest].[dbo].[ProfilerResults]
ORDER BY ApplicationName |
| Transactions Per Database |
Most Used Objects |
-- Transaction per database in descending order
SELECT COUNT(DatabaseID), DB_NAME(DatabaseID)
FROM [ProfilerTest].[dbo].[ProfilerResults]
GROUP BY DatabaseID
ORDER BY DatabaseID DESC |
-- Most used objects in descending order
SELECT COUNT(ObjectID) AS 'ObjectCount', OBJECT_NAME(ObjectID) AS 'ObjectName'
FROM [ProfilerTest].[dbo].[ProfilerResults]
GROUP BY ObjectID
ORDER BY ObjectCount DESC |
Note: In the code snippets above, replace the FROM clause, i.e., [ProfilerTest].[dbo].[ProfilerResults] with the database, schema and table where you captured your results.
Based on your analysis, apply the following rules to improve the query performance:
- Review the query plan for possible improvements in queries that consistently appear as offenders in Profiler analyses.
- Leverage the Database Engine Tuning Advisor for index and partitioning recommendations.
- Review the query join order to minimize the number of rows that are being processed.
- Validate that indexes meet individual query needs for the JOIN, WHERE, GROUP BY and ORDER BY clauses.
- Ensure that data types are the most efficient for the data (i.e., use integers for keys as opposed to character data types).
- Remove any unneeded logic in the statements such as GROUP BY or ORDER BY when they are not needed by users.
- Return only the needed amount of data and nothing more.
- Ensure statements do not explicitly hold any transactions longer than necessary.
Starting off, you may need to chip away at the statements one at a time, but over time you will begin to find recognizable patterns with similar solutions yielding a significant performance improvement.
Top 5 SQL Server query tips
Home: Introduction
Tip 1: SQL Server query design: 10 mistakes to avoid
Tip 2: Troubleshoot SQL Server queries and improve I/O
Tip 3: Tracking query execution with SQL Server 2005 Profiler
Tip 4: Find and fix resource-intensive SQL Server queries
Tip 5: Running analytical queries with Analysis Services
For more help, check out our FAQ: SQL Server query errors and explanations.
About the author: Jeremy Kadlec is the principal database engineer at Edgewood Solutions, a technology services company delivering professional services and product solutions for Microsoft SQL Server. He has authored numerous articles and delivers frequent presentations at regional SQL Server users groups and nationally at SQL PASS. Kadlec is the SearchSQLServer.com Performance Tuning expert. Ask him a question here.
More information from SearchSQLServer.com
Tip: SQL Profiler: Features, functions and setup in SQL Server 2005
Tip: Using SQL Server Profiler to capture a performance baseline
Topic: Research performance and tuning best practices
SQL Server Profiler results
