SQL Performance Turning
Peter Gulutzan, Trudy Pelzer · 2002
From the Book: A poorly performing database application can cost each user time, and have an impact on other applications running on the same computer or the same network. The purpose of this book is to help you improve the performance of your SQL database. It is not an introduction to SQL syntax, not a tuning manual for a specific SQL implementation, and not a guide to design or data modelling. Instead, we've written it for users and programmers who want to improve SQL performance, no matter what brand of SQL they use. In this book, you'll find tuning tips for common situations. Topics that will give you the flavor include: How to change a query so it will go faster, What an index does, and Shifting work from the server to the client. Rather than exploiting the unique features of a single DBMS, we're going to give you ideas that are good for all of the major SQL DBMSs. Client/server programmers and consultants need to appreciate what can happen when the DBMS changes, orthe most challenging situationthe DBMS is unknown. So we tested all the ideas in this book on eight well-known DBMSs. A good DBMS already contains a good optimizer. Yet you have picked up a book that promises to help you do your own tuning. That means that either you don't know something ... or you do know something: You don't know that your DBMS is good. You know that even good tools work better in skilled hands. You don't know that your DBMS is good ... That would be true if you're a student, or new on the job. That would especially be true if you're writing queries or programs that should work on more than one DBMS. You're most likely toencounter one of the DBMSs with the largest market shares (based on figures for the year 2000): Oracle 33.8% IBM 30.1% (prior to acquisition of Informix) Microsoft 14.9% Sybase 3.2% Informix 3.0% Data from: Gartner Dataquest (May 2001) You might also run into DBMSs that are popular for web work (like MySQL), for work with Borland products (like InterBase), for desktops (like Access), for mobile and Java interfacing (like Cloudscape), for embedded systems (like Solid), or a host of small fry (like mSQL and gadfly). This book tries to be useful for the common denominator in all products. To use automotive analogies, it's not a mechanic's guide to tuning the 1999 Cadillac book, it's a driver's guide to performance of modern cars book—even if you have a manual transmission. You know that even good tools work better in skilled hands ... Everybody has heard of sluggish SQL queries, or even whole systems, that some hero(ine) improved with small effort. Usually the improvement is small too, so we will avoid extravagant promises. But we will make you the following guarantees: You will be able to follow our arguments without deep thinking or hard work. All we assume is that you have basic knowledge of programming and standard SQL syntax. In fact, you can read this book on a plane or at the beach. All of our observations have been confirmed by tests on real DBMSs within the last several months. We know that optimizing is very different from over-clocking and will only discu techniques. One accusation that could be levelled, and to which we plead guilty, is that some of our material is ad-hoc instead of general principles. Of course! There are only a few general principles in DBMS optimization: First, do no harm. (Actually that's from Hippocrates' ancient manual on medical treatments. It applies to anybody fixing what ain't really broke.) Get more memory, add indexes, re-cable the network. (If you can influence the environment like that, then do so.) Design right in the first place. Understand the query. Instead of general principals, we will be looking at what can be done with what's likely to be at hand. If we descend sometimes to mere tips and warnings about traps, that's because we've seen over the years that examples of real situations can help people realize what the general rules are. As has often been observed, tips should be based on principles. The DBMSs that we looked at while preparing this book include IBM DB2, Informix, Ingres II, InterBase, Microsoft SQL Server, MySQL, Oracle, and Sybase (MS-WindowsNT versions). Each was installed and tested using the default switches recommended in the vendors' instructions. To avoid favoring any vendor's idiosyncrasies, all SQL examples in this book are written in ANSI/ISO Standard SQL:1999. Host-language examples are written in C plus ODBC, or Java plus JDBC. Your DBMS is your pal. We won't counsel you to subvert it, be paranoid about it, or insult it by assuming it's stupid. Rather, as you would with a pal, you should try to get to know it better, and help it to help you.