Book Rookery

In this installment of the Book Rookery, High Performance MySQL authors Jeremy Zawodny and Derek J. Balling share some of the MySQL lessons they've learned over the years and offer insight into the performance gains possible when you use the techniques covered in the book.

What kinds of applications do people use MySQL for? Is it really free?

People are using MySQL in all types of applications, both personal and corporate, from "quickie one-off applications" to helping power Sabre's ticketing reservation system.

For almost all users, it really is free, since it is released under the GPL. There has been some confusion about licensing issues, namely commercial versus noncommercial use and what constitutes "redistribution." Their Web site likes to make the analogy "If you make money, you need a license," which is simply not the case, since the GPL'ed code is free for use in both commercial and noncommercial applications. The only time it would not be free is if you intend to redistribute code based on the MySQL code, and don't intend to release your own code as open source. In that case, you would need to purchase a commercial license from MySQL AB, to permit you to redistribute binary-only versions of your derivative work. For almost all Web-based applications, though, that restriction is moot since you almost never redistribute your code to an outside party.

And, in reality, most companies will buy a support agreement for the software, which helps to ensure they get help when needed and MySQL AB is able to stay in business.

Can you tell us more about the intention of the book and what's covered inside?

Jeremy first conceived the book when he was encountering growth problems while deploying MySQL at Yahoo. They weren't necessarily "MySQL problems," per se, but they were gotchas and best-practice configurations that hadn't really been documented anywhere (or at least not very well). Our book was conceived as a place to write down the lessons learned over time, and try to put them all in one place so that future administrators didn't have to scour the Web, or mailing lists, or even worse, try to pick through the source code, in order to find the answer to their problem.

We cover a wide range of topics, starting from baseline decisions like "what storage engine should I use for my data," and progressing further into query optimization, hardware configurations, replication, and load balancing. We also touch on, because it's important, the security and backup situations that a large installation will encounter.

How much of a performance increase do you think you can make through using the techniques outlined in your book?

That depends a great deal on what you're starting out with in the first place. If you've got a fairly well-designed database, on decent hardware, maybe you don't see much improvement at all. On the other hand, if you're like many installations where MySQL made its inroads "through the back door," and there's not been a lot of formal DBA experience in the organization, it's possible the optimizations we discuss can give you performance increases of several thousand percent.

If you could only tweak one system within MySQL to get the best performance, what would you tweak?

The key_buffer setting for the MyISAM storage engine. Once you set up the correct indexes, MySQL needs sufficient memory to keep the most actively used indexes cached in memory. For primarily InnoDB users, the answer is the innodb_buffer_pool for very similar reasons.

You must have experienced lots of different databases of information over the years - what was your favorite use for a database system?

That's a tough one. I (Jeremy) really like the aviation database that Jeremy Cole (MySQL AB's training manager) has been building. He's combined freely available information from the FAA and NTSB with MySQL and a simple Web interface in a way that brings together previously separate and hard-to-find information. If you want to know which airports or airlines experience the most delays, it'll tell you. If you want to know what types of planes American Eagle flies, it'll tell you.

What is the relative performance using the various underlying database technologies with MySQL (InnoDB vs ISAM, etc.)?

That's going to depend a great deal on the type of data you have, how you access it (write once read often? write often read seldom?), and how complex your needs are (for example, do you need transactions, or are you just simply using the database as your Apache server log?).

Do you believe that MySQL can compete with more commercial systems, like SQL Server?

Over time, yes. MySQL doesn't have all the bells and whistles that SQL Serve, Oracle, or DB2 have. But what it lacks in features it generally makes up for in simplicity and raw performance, both of which often translate into cost savings.

As times goes on, MySQL will get closer and closer to the "big name" databases.

How much of a performance hit is there in using a remote database versus a local one via sockets?

If the "remote" database is fairly close, on a network-scale the performance impact is pretty marginal (a few milliseconds). Most of the bottlenecks in MySQL (or any database, really) are in disk accesses - how quickly can the database engine find the right spot on the disk to give you the information you asked for. Almost everything you optimize in a database server centers around speeding up that basic act. The added time for accessing a database via TCP/IP instead of via a Unix socket, is fairly negligible. If you have the server truly "remote" (like, the other side of the country), then you might have issues with network latency.

What is your favorite cartoon daily?

Derek says: Depending on my mood, either "Dilbert" or "Doonesbury"... I guess it just depends on which is frustrating me more lately, work or politics.

Jeremy says: "Dilbert" or "The Far Side," which is sadly no longer being written.

About the Authors
Jeremy Zawodny is Yahoo's resident MySQL Geek and the lead author of High Performance MySQL. He lives in San Jose, California and flies gliders around Northern California and Nevada in his spare time. He also maintains a Weblog at http://jeremy.zawodny.com/blog.

Derek J. Balling has been a Perl programmer and Unix/Linux system administrator since 1996, having helped build two different ISPs from the ground up in the midwestern United States. He spent several years of his career at Yahoo!, working in their Infrastructure Group, where he worked on tools to help improve system uptime. He presently works at a healthcare supply company, helping infiltrate the open source virus into their infrastructure.