Abstract
Web applications must be load tested to analyze their behavior under various load conditions. Typically, these load tests are automated using protocol-level HTTP requests (e.g., using JMETER). However, there are several disadvantages to using protocol-level requests for load tests. For example, protocol-level requests are only partially representative of the true usage of a web application, as the web application is not actually executed in a browser. It can be difficult to abstract complex behavior, such as a login sequence, into requests without executing the application. Browser-based load testing can be used as an alternative to protocol-level requests. Using a browser-based testing framework, such as SELENIUM, tests can be executed more realistically - inside a browser. Unfortunately, because a browser instance must be started to conduct a test, browser-based testing has a high performance overhead which limits its applicability for load tests. In this paper, we propose an approach for reducing the performance overhead of running SELENIUM-based load tests. Our approach shares browser instances between test user instances, thereby reducing the performance overhead that is introduced by launching many browser instances during the execution of a test. Our experimental results show that our approach can significantly increase the number of user instances that can be tested on a test machine without overloading the load driver. Our approach and the experiences that we share in this paper can help software practitioners improve the efficiency of their own SELENIUM-based load tests.