Summary

Load testing is an important long-term function that is used to assist management in understanding the behavior of their systems. Of particular importance is the inflection point, where the relationship between applied load and system response shifts from being linear to nonlinear. It can be used in setting operations alerts and parameters as well as in anticipating future problems caused by lack of resources. Load testing can also be used to help build models, as discussed in Chapter 12.

Load testing on the Web has some characteristics that are quite different from load testing in pre-Web, transaction-oriented systems.

The first key difference is the appearance of flash load. This occurs because external users connecting from the public Internet can appear in unprecedented numbersmuch greater than seen in controlled, proprietary systems.

The second key difference is abandonment and the fact that Web systems usually cannot detect abandonment directly. Unlike call center operators and employees using an intranet, the Internet's web users quickly abandon a transaction if the response time is too long. That abandonment affects system load and therefore the response time for other users. Also, as the system usually cannot directly detect abandonment but must use timeouts instead, abandoned transactions can congest the system if resources are not efficiently recovered.

Load testing on the Web must therefore include user abandonment behavior and endurance tests to evaluate the system's ability to detect abandoned transactions and recover resources efficiently.

Of course, Web load testing must also involve all Web services, networks, and other equipment to ensure that it is a realistic view of actual performance. Web load test services may be useful to provide large-scale, highly realistic load testing as a final step before production.