The Internet has over 1 billion users. If your web site attracts just one tenth of one percent of them each month and they look at 10 pages, your servers may need to handle sustained loads over 40,000 hits an hour.

For read-only web applications, performance testing is simple. Any number of tools can be used to simulate load on a server using GET and following links. For transactional systems, things are not so simple. Scripts need to move items through workflows, enter data that passes validation, check results, and in general, behave more like thousands of fast clicking users rather than a simple web crawler.

You need to consider:

• existing page hit analysis
• the number of simultaneous requests needed for simulations
• flexible data strategies
• fast turn-around times
• verifying simulations match expectations
• scaling with distributed load generation

This talks looks at strategies for determining what you should measure before you put you next release online, how to go about it, and why it's less work to build some simple tools with Ruby than it is to wrestle with expensive vendor products. We also describe the some of the solutions we implemented to get more performance out of our Rails application.