There are very few applications that don't behave differently when they come under load. To be able to investigate how the system will react when put under expected or even extreme load, we use load generators.

It's nearly impossible to mimic all the different use cases and their individual usage. Therefore, the load tests are often limited to 10 or 20 selected use cases. Some of them should be the most common ones, and some known or suspected trouble makers.

It's often tempting to limit the tests to read-only ones to minimize the trouble of setting up a database with a known state before each execution. This may be okay in some cases, but be prepared to add read-write tests at some stage before entering production in order to reach acceptable levels of QoS.

Extensive load testing may be limited by the chosen load-generator software and the operating system, network, and so on in such a way that it can't generate the load needed. The solution is to add more distributable "slave...