Web Performance Tuning: Speeding up the Web [Paperback]

The thing is, if you need this book, your website is probably a high-traffic professional/commercial site. And in these days this means (1) dynamic content, (2) database, (3) a content-management/templating system, (4) user identity tracking. Perhaps even interface to legacy client/server systems. Unfortunately, this book goes only as far as CGI, Java, and general DB issues. Messaging middleware is briefly considered. Distributed OO (CORBA, EJB) is discussed and dismissed (a luxury in real world). No coverage of other popular dynamic web technologies (e.g. ASP, ColdFusion) or content-management systems. In particular, a serious discussion of trade-offs between performance and content/workflow manageability would ground the whole discussion in real life.

And the architecture chapter, while very insightful, is simply too thin. After all it is much better and easier to plan for performance from the start, then to try tweaking an existing system. The chapter discusses architectures of varying complexity - <i>without including a single diagram!</i> Complete case studies along the line of the mod_perl white paper .... would be invaluable - perhaps broken down by type (e.g. news/portal/B2C) where unique usage patterns will drive unique architecture and optimization.

Despite the tilt towards monitoring and diagnosis, this is still a very valuable book in an under-served but important area. Generous references enable the reader to explore individual topics further.

The book's first section, Preliminary Considerations, is an outstanding analysis of the relationships between bandwidth, latency, server memory, CPU speed, traffic levels, user expectations and cost. Along the way, the author highlights the extreme gap between real-world performance requirements and the artificial numbers generated by benchmark tools. He notes that a full T1 line can only carry 33 hits per second (at 4K each), and that a million hits per day translates into a peak server load of only about 25-30 hits per second. These real-world numbers are then contrasted with the hundreds or thousands of hits per second usually quoted by vendors, which the author refers to as "benchmarketing." Refreshingly, the author then describes how to create practical benchmark scenarios for your own Web servers, and how to use them effectively.

The second section, Tuning In Depth, briefly discusses Web client tuning, and then addresses the details of network, Web server, and CGI tuning. The author explains each issue, makes specific recommendations, and supports them with relevant facts and calculations. Each chapter ends with a concise "key recommendations" section, which condenses the chapter into a few memorable one-liners - a great feature for the busy Webmaster. The recommendations run from very general guidelines to very specific suggestions, such as "Use separate disks for log writing and content reading." While some of the discussion applies only to UNIX servers, most of the recommendations apply equally well to other platforms.

Finally, the book includes Appendixes with specific tuning tips for Netscape Enterprise Server, Apache, and Solaris' 2.x TCP/IP Stack. Although much of the same material is available on the Web (with updates), the printed reference and the author's comments are valuable resources to have handy if you use these products.

This book should be considered required reading for all present and future Webmasters; it is the most clear and direct discussion of real-world Web server performance published to date. However, this book's UNIX-centric view skips over some important issues facing today's Webmasters, such as Web database performance and the tuning of non-UNIX Web servers. The book does not mention FileMaker or Access, or middleware products like Tango, Lasso, or Cold Fusion. And while the tuning guidelines will be helpful to most Webmasters, the book does not provide any specifics for optimizing Microsoft IIS or WebSTAR. It is a bit surprising to see all of these popular packages omitted from this very recent book. Ultimately though, every Webmaster who reads this book will learn new ways to improve server performance and many of them will enjoy it as well.

As the web is changing every day some of the information is dated, especially the chapters on running server side applications. The chapter on CGI is decent, but the chapters on database and Java tuning are cursory and best covered by books dedicated to those subjects. There is nothing on active server pages. Also a chapter on balancing security versus performance would have been welcome, and hopefully will be included in a second edition.

There is definitely more about UNIX than NT in the book. This doesn't matter when doing hardware and network tuning and Microsoft certainly does not help with their license restriction on the publication of IIS benchmarks. The reality is that there are more web servers running UNIX or Linux variants than NT. However, with the rapid proliferation of active server pages more should be included on NT in a future edition.

Getting usable information on performance tuning is sometimes very difficult. Such information is usually gleaned sparingly from Usenet groups or expensively from consultants. "Web Performance Tuning" is a solid guide with a lot of information condensed and indexed that would be difficult to find elsewhere. It is definitely remaining on the easy to reach side of my bookcase.