Table of Contents
- Microsoft SQL Server Defined
- Microsoft SQL Server Features
- Microsoft SQL Server Administration
- Microsoft SQL Server Programming
- Performance Tuning SQL Server: Tools and Processes
- Performance Tuning SQL Server: Tools Overview
- Creating a Performance Tuning Audit - Defining Components
- Creating a Performance Tuning Audit - Evaluation Part One
- Creating a Performance Tuning Audit - Evaluation Part Two
- Creating a Performance Tuning Audit - Interpretation
- Creating a Performance Tuning Audit - Developing an Action Plan
- Understanding SQL Server Query Plans
- Performance Tuning: Implementing Indexes
- Performance Monitoring Tools: Windows 2008 (and Higher) Server Utilities, Part 1
- Performance Monitoring Tools: Windows 2008 (and Higher) Server Utilities, Part 2
- Performance Monitoring Tools: Windows System Monitor
- Performance Monitoring Tools: Logging with System Monitor
- Performance Monitoring Tools: User Defined Counters
- General Transact-SQL (T-SQL) Performance Tuning, Part 1
- General Transact-SQL (T-SQL) Performance Tuning, Part 2
- General Transact-SQL (T-SQL) Performance Tuning, Part 3
- Performance Monitoring Tools: An Introduction to SQL Profiler
- Performance Tuning: Introduction to Indexes
- Performance Monitoring Tools: SQL Server 2000 Index Tuning Wizard
- Performance Monitoring Tools: SQL Server 2005 Database Tuning Advisor
- Performance Monitoring Tools: SQL Server Management Studio Reports
- Performance Monitoring Tools: SQL Server 2008 Activity Monitor
- The SQL Server 2008 Management Data Warehouse and Data Collector
- Performance Monitoring Tools: Evaluating Wait States with PowerShell and Excel
- Practical Applications
- Professional Development
- Application Architecture Assessments
- Business Intelligence
- Tips and Troubleshooting
- Additional Resources
General Transact-SQL (T-SQL) Performance Tuning, Part 3
Last updated Mar 28, 2003.
I’m in a series of tutorials that explains how to tune the queries you have within your system. If you haven’t read the first two tutorials, it’s a good idea to stop here and take care of that now. Keep in mind as you read through this series that tuning the queries in your system is only part of the picture. There are other components for tuning the entire system, and I’ve covered that process in this section of the Guide.
In the first installment of this series, I explained that “knowledge is power.” I explained how to set up your environment so that you can observe the queries without interference from other queries running on the system. This is similar to “unit testing”, which is where you test that what you’re doing works, and works as expected.
Of course, this “isolation” performance testing is useful, but in production you’ll be faced with more activity than just one query — this is where locking and blocking begins to take effect. I’ve explained this behavior before, and you can check out those articles for how to deal with them.
In the second part I explained the primary tool that you can use to examine your queries as they run on the server, and a little about how to interpret the results. At the very least you should examine the Query Plan for your statements to ensure they are as fast as they can be.
In this tutorial, I’ll explain a few tips for speeding up the queries by avoiding the problems that you’ll find with the knowledge and the tools that I showed you in the previous two installments. I’ll explain the primary issues I’ve seen with queries in my years of experience. I won’t cover every issue, but if you make sure you don’t commit these errors (or at least correct them if you do), you can really speed up your system.
Getting What You Don’t Need
Of all the query tuning advice I can give, the issue I’ve seen most often and the most basic problem that I see ignored even by professionals is getting more data in a query than the requester really wants.
This problem is so easy to create if you don’t take the entire environment in mind. You create a view, a Stored Procedure or some other query instrument that requests a set of fields from one or more tables. When you first design the query, you might be tempted to create a view or Stored Procedure that can be used for a lot of purposes - but this is a mistake.
Let’s say you have a need to show visit information for a veterinarian. You might be tempted to create a view that pulls back the doctor’s information, the animal’s information, the information about the trip, the medical supplies used and so on. You could then use this view for a full report, or perhaps just to get info on a doctor, or an animal and so on. The point is that when you over-use a broad query, you’re asking the system to bring back more data than you need. This also holds true for a Stored Procedure.
I’ve also seen far too much use of SELECT *. There is almost never a good reason to do this. If you have this in your applications, take it out and select the columns you need. In fact, it’s often better to select all columns by name than to use the asterisk because of the processing the engine has to do in some cases to determine the column names.
Developers will sometimes get a little lazy about this and just copy the SELECT statements from one form to another, without realizing that thousands and thousands of requests might be in store for the application, all getting columns not needed for a particular form.
The Fix: Although this is the easiest and most common mistake to make, it can be a little difficult to fix after the fact. But it can be done. Just examine the queries using Profiler, paying very close attention to any queries that might use SELECT *. Track those down and find out why — if you can fix them without breaking an application, do.
For your own queries, make more views, Transact-SQL statements or Stored Procedures if you need different data for a specific purpose. It might seem like more work up front, but you will reap the performance benefits if you do.
Also, make sure you spend some time educating your team to only get what they need.
Incorrect or Non-Existent Indexes
The next most common problem is using, or not using, an index. While that might sound a little contradictory, it isn’t. To find out why, you need to think back to what an Index does. It is simply another table (Clustered indexes are another matter that I’ll explain in a moment) that points to entries in a table. They are remarkably similar to an index in the back of a book.
Indexes are very useful for read operations. They provide a pointer to the location in the original table, making it very easy to find data. But they are only useful if they are up to date - meaning that they need to be reorganized during standard maintenance.
So if you have a query that references the same columns over and over, you should evaluate it for an index. You’ll see lots of “Table Scan” operations in the Query Processor, and you’ll see that most of the time spent on the query is there.
But you actually pay a penalty on those indexes when you update or insert data. Those indexes now need to point to the new data, and if the index entry doesn’t fit on the 8K “Page” of storage where it lives, a physical operation takes place to move the index data around.
This is really an issue with clustered indexes. Clustered indexes aren’t in fact another table at all — the “base” table gets physically arranged on disk in the order specified in the index statement. That’s great for values that always increase and are unique, because the table just adds entries to the end of the data storage. However, if you’ve chosen a column to physically arrange the data on disk that changes often or has inserts above the location of the last entry, all that data has to “move over” to allow the new entry. That’s a very expensive set of physical operations, and as usual the physical operations are a lot slower than doing things in memory or on the CPU.
The Fix: Use the dynamic management views in SQL Server 2005 and higher to ensure the indexes are really being used. If not, drop them and monitor the system to ensure there isn’t a problem with doing that.
Use the various query evaluation tools to ensure that you have indexes where needed. If the table experiences lots of reads and not many writes, consider adding an index. Then test.
Carefully evaluate your clustered indexes. If they are not on a sequentially increasing value, such as a primary key, then consider changing it. By the way, primary and foreign keys almost always need an index.
Out of Date Statistics
Statistics are another structure that SQL Server uses to make query access faster. In essence, they are another type of index, but they are used differently. They help SQL Server decide whether an index is useful, along with other information needed to make the query run quickly.
The Fix: You should either use the “Auto Update” option on statistics, or update them manually or a regular schedule. Normally this is part of standard maintenance.
Incorrect Use of Triggers
Triggers are great - except when you don’t think them through. Triggers are code placed on INSERT, UPDATE or DELETE operations that “fire” whenever one of these happen to a table. If you have a lot of logic in these triggers, or they hit other tables that have triggers, you can cause an intense amount of activity on your system, especially if there are a lot of tables affected.
The other issue with triggers is that they are often “invisible” - since they don’t show up in code easily, all the developer knows is that something massive happens every time there is an INSERT, UPDATE or DELETE.
The Fix: Examine each trigger to see what they do - is this something that might be handled in a stored procedure instead? Would it be something you could move into the program itself?
Functions and Computed Columns
You can create a User Defined Function to make a column do math or other operation to make a “derived” column value. This can be very expensive, if the calculations are intense. This is magnified if you have the very first problem, selecting more data than you need. Each and every time you’re performing work you may not need.
The Fix: Once again, evaluate whether this is really needed. Could you process this on the application side better, or perhaps even store the data rather than calculate it?
As you can see, all three of these tutorials work together. Knowing what your application needs to do, evaluating it using the tools SQL Server provides, and then checking to make sure you perform the operations correctly will help you speed your applications along. The best way to tune a system is to minimize the work it needs to do to answer the queries.
InformIT Articles and Sample Chapters
If you’ve come into this tutorial without reading the complete series I have on performance tuning, you can find that here.
Books and eBooks
You can’t do better than this work, SQL Performance Tuning, by Peter Gulutzan and Trudy Pelzer.